refactor(ai-proxy): replace per-protocol request_body override with provider capability hooks

Replace the per-protocol keyed request_body override design with a simpler
flat schema where users set max_tokens and APISIX automatically maps it to
the correct field name via rewrite_request_body hooks in provider capabilities.

Config changes from:
  request_body: { "openai-chat": { max_tokens: 500 } }
to:
  request_body: { max_tokens: 500 }

Each provider's capability entry now has a rewrite_request_body(body, override, force)
hook that sets the provider-native field:
- OpenAI: max_completion_tokens
- OpenAI Responses: max_output_tokens
- Gemini/Vertex-AI: max_completion_tokens
- DeepSeek/Anthropic/OpenRouter/AIMLAPI/Azure/Compatible: max_tokens

Removed:
- merge.lua (deep merge no longer needed)
- protocols.names() (no longer needed by schema)

Author: nic-6443

apisix/plugins/ai-protocols/init.lua

@@ -22,7 +22,6 @@
 
 local converters = require("apisix.plugins.ai-protocols.converters")
 local ipairs = ipairs
-local pairs = pairs
 
 local _M = {}
 
@@ -66,16 +65,6 @@ function _M.get(name)
 end
 
 
---- Get the list of all registered protocol names.
--- @return table Array of protocol names
-function _M.names()
-    local names = {}
-    for name in pairs(registered) do
-        names[#names + 1] = name
-    end
-    return names
-end
-
 
 --- Find a converter that can bridge from client_protocol to a protocol
 -- supported by the driver. Delegates to the converters registry.

apisix/plugins/ai-providers/aimlapi.lua

@@ -15,12 +15,23 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(body, override, force)
+    if override.max_tokens then
+        if force or body.max_tokens == nil then
+            body.max_tokens = override.max_tokens
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new(
     {
         host = "api.aimlapi.com",
         port = 443,
         capabilities = {
-            ["openai-chat"] = { path = "/chat/completions" },
+            ["openai-chat"] = {
+                path = "/chat/completions",
+                rewrite_request_body = rewrite_max_tokens,
+            },
         },
     }
 )

apisix/plugins/ai-providers/anthropic.lua

@@ -15,13 +15,27 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(body, override, force)
+    if override.max_tokens then
+        if force or body.max_tokens == nil then
+            body.max_tokens = override.max_tokens
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new(
     {
         host = "api.anthropic.com",
         port = 443,
         capabilities = {
-            ["openai-chat"] = { path = "/v1/chat/completions" },
-            ["anthropic-messages"] = { path = "/v1/messages" },
+            ["openai-chat"] = {
+                path = "/v1/chat/completions",
+                rewrite_request_body = rewrite_max_tokens,
+            },
+            ["anthropic-messages"] = {
+                path = "/v1/messages",
+                rewrite_request_body = rewrite_max_tokens,
+            },
         },
     }
 )

apisix/plugins/ai-providers/azure-openai.lua

@@ -15,12 +15,23 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(body, override, force)
+    if override.max_tokens then
+        if force or body.max_tokens == nil then
+            body.max_tokens = override.max_tokens
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new(
     {
         port = 443,
         remove_model = true,
         capabilities = {
-            ["openai-chat"] = { path = "/completions" },
+            ["openai-chat"] = {
+                path = "/completions",
+                rewrite_request_body = rewrite_max_tokens,
+            },
         },
     }
 )

apisix/plugins/ai-providers/base.lua

@@ -36,7 +36,6 @@ local transport_http = require("apisix.plugins.ai-transport.http")
 local transport_auth = require("apisix.plugins.ai-transport.auth")
 local log_sanitize = require("apisix.utils.log-sanitize")
 local protocols = require("apisix.plugins.ai-protocols")
-local deep_merge = require("apisix.plugins.ai-proxy.merge").deep_merge
 local ngx = ngx
 local ngx_now = ngx.now
 
@@ -183,14 +182,12 @@ function _M.build_request(self, ctx, conf, request_body, opts)
         end
     end
 
-    -- Inject per-target-protocol request body override (deep merge)
-    if opts.request_body_override_map then
-        local patch = opts.request_body_override_map[ctx.ai_target_protocol]
-        if patch then
-            core.log.info("applying request_body override for target protocol '",
-                          ctx.ai_target_protocol, "'")
-            request_body = deep_merge(request_body, patch,
-                                      opts.request_body_force_override)
+    -- Apply request body override via provider capability hook
+    if opts.override_request_body then
+        local cap = self.capabilities and self.capabilities[ctx.ai_target_protocol]
+        if cap and cap.rewrite_request_body then
+            cap.rewrite_request_body(request_body, opts.override_request_body,
+                                     opts.request_body_force_override)
         end
     end
     params.body = request_body

apisix/plugins/ai-providers/deepseek.lua

@@ -15,12 +15,23 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(body, override, force)
+    if override.max_tokens then
+        if force or body.max_tokens == nil then
+            body.max_tokens = override.max_tokens
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new(
     {
         host = "api.deepseek.com",
         port = 443,
         capabilities = {
-            ["openai-chat"] = { path = "/chat/completions" },
+            ["openai-chat"] = {
+                path = "/chat/completions",
+                rewrite_request_body = rewrite_max_tokens,
+            },
         },
     }
 )

apisix/plugins/ai-providers/gemini.lua

@@ -15,12 +15,23 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(body, override, force)
+    if override.max_tokens then
+        if force or body.max_completion_tokens == nil then
+            body.max_completion_tokens = override.max_tokens
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new(
     {
         host = "generativelanguage.googleapis.com",
         port = 443,
         capabilities = {
-            ["openai-chat"] = { path = "/v1beta/openai/chat/completions" },
+            ["openai-chat"] = {
+                path = "/v1beta/openai/chat/completions",
+                rewrite_request_body = rewrite_max_tokens,
+            },
         },
     }
 )

apisix/plugins/ai-providers/openai-compatible.lua

@@ -15,10 +15,26 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(field_name)
+    return function(body, override, force)
+        if override.max_tokens then
+            if force or body[field_name] == nil then
+                body[field_name] = override.max_tokens
+            end
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new({
     capabilities = {
-        ["openai-chat"]       = { path = "/v1/chat/completions" },
-        ["openai-responses"]  = { path = "/v1/responses" },
+        ["openai-chat"]       = {
+            path = "/v1/chat/completions",
+            rewrite_request_body = rewrite_max_tokens("max_tokens"),
+        },
+        ["openai-responses"]  = {
+            path = "/v1/responses",
+            rewrite_request_body = rewrite_max_tokens("max_output_tokens"),
+        },
         ["openai-embeddings"] = { path = "/v1/embeddings" },
     },
 })

apisix/plugins/ai-providers/openai.lua

@@ -15,13 +15,29 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(field_name)
+    return function(body, override, force)
+        if override.max_tokens then
+            if force or body[field_name] == nil then
+                body[field_name] = override.max_tokens
+            end
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new(
     {
         host = "api.openai.com",
         port = 443,
         capabilities = {
-            ["openai-chat"]       = { path = "/v1/chat/completions" },
-            ["openai-responses"]  = { path = "/v1/responses" },
+            ["openai-chat"]       = {
+                path = "/v1/chat/completions",
+                rewrite_request_body = rewrite_max_tokens("max_completion_tokens"),
+            },
+            ["openai-responses"]  = {
+                path = "/v1/responses",
+                rewrite_request_body = rewrite_max_tokens("max_output_tokens"),
+            },
             ["openai-embeddings"] = { path = "/v1/embeddings" },
         },
     }

apisix/plugins/ai-providers/openrouter.lua

@@ -15,12 +15,23 @@
 -- limitations under the License.
 --
 
+local function rewrite_max_tokens(body, override, force)
+    if override.max_tokens then
+        if force or body.max_tokens == nil then
+            body.max_tokens = override.max_tokens
+        end
+    end
+end
+
 return require("apisix.plugins.ai-providers.base").new(
     {
         host = "openrouter.ai",
         port = 443,
         capabilities = {
-            ["openai-chat"] = { path = "/api/v1/chat/completions" },
+            ["openai-chat"] = {
+                path = "/api/v1/chat/completions",
+                rewrite_request_body = rewrite_max_tokens,
+            },
         },
     }
 )