fix: [AI-6622] enable cache_control trigger for altimate-backend provider (#850)

Adds `model.providerID === "altimate-backend"` to the cache-control gating
condition in `ProviderTransform.message()` so messages routed through the
Altimate Gateway get `cache_control: ephemeral` markers stamped on the last
content block. Required for the gateway's Anthropic-fallback path: when
Azure GPT-5.5 errors, litellm falls back to Anthropic Sonnet 4.6, which
honors these markers; without them, the fallback calls would not cache.

For the Azure-primary path: Azure ignores `cache_control` markers (it uses
automatic prefix-based caching), so stamping them is a no-op there. The
companion backend change extracts `prompt_tokens_details.cached_tokens` from
the response and surfaces it back to clients so cache rates and per-session
cost become visible.

Companion backend change on `altimate-backend` branch
`fix/AI-6622-llm-cache-plumbing`. Both repos pair together.

Full RCA + investigation journey: `~/Documents/altimate-deliverables/AI-6622-*.md`.

Co-authored-by: Haider <haider@altimate.ai>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: 3 people

packages/opencode/src/provider/transform.ts

@@ -284,6 +284,7 @@ export namespace ProviderTransform {
     if (
       (model.providerID === "anthropic" ||
         model.providerID === "google-vertex-anthropic" ||
+        model.providerID === "altimate-backend" ||
         model.api.id.includes("anthropic") ||
         model.api.id.includes("claude") ||
         model.id.includes("anthropic") ||