Fix build: forward-declare MultimodalInput in irunner.h

irunner is a header-only Buck target with no deps, so it cannot
include multimodal_input.h (owned by multimodal_runner_lib which
requires aten-variant deps). Use a forward declaration instead.

Move the string convenience prefill(prompt, num_bos, num_eos) from
IRunner to TextLLMRunner and MultimodalRunner, where
multimodal_input.h is available for constructing MultimodalInput.

This PR was authored with the assistance of Claude.

Author: kirklandsign

extension/llm/runner/irunner.h

@@ -16,14 +16,15 @@
 #include <string>
 #include <vector>
 
-#include <executorch/extension/llm/runner/multimodal_input.h>
 #include <executorch/extension/llm/runner/stats.h>
 #include <executorch/runtime/core/error.h>
 
 namespace executorch {
 namespace extension {
 namespace llm {
 
+class MultimodalInput; // Forward declaration
+
 // Configuration struct for generation parameters, fields should be sorted in
 // alphabetic order
 struct GenerationConfig {
@@ -146,16 +147,6 @@ class ET_EXPERIMENTAL IRunner {
     return runtime::Error::NotSupported;
   }
 
-  /**
-   * Convenience overload: prefill a single text prompt.
-   */
-  runtime::Error
-  prefill(const std::string& prompt, int32_t num_bos = 0, int32_t num_eos = 0) {
-    std::vector<MultimodalInput> inputs;
-    inputs.emplace_back(MultimodalInput(prompt));
-    return prefill(inputs, num_bos, num_eos);
-  }
-
   /**
    * Stop the generation process.
    */

extension/llm/runner/multimodal_runner.h

@@ -138,8 +138,6 @@ class ET_EXPERIMENTAL MultimodalRunner : public IRunner {
       std::function<void(const std::string&)> token_callback = {},
       std::function<void(const Stats&)> stats_callback = {});
 
-  using IRunner::prefill; // Bring in string convenience overload
-
   /**
    * Prefill multimodal inputs to fill the KV cache, for example to reload
    * chat history. Call generate() with a non-empty prompt afterwards to
@@ -155,6 +153,16 @@ class ET_EXPERIMENTAL MultimodalRunner : public IRunner {
       int32_t num_bos = 0,
       int32_t num_eos = 0) override;
 
+  /**
+   * Convenience overload: prefill a single text prompt.
+   */
+  ::executorch::runtime::Error
+  prefill(const std::string& prompt, int32_t num_bos = 0, int32_t num_eos = 0) {
+    std::vector<MultimodalInput> inputs;
+    inputs.emplace_back(MultimodalInput(prompt));
+    return prefill(inputs, num_bos, num_eos);
+  }
+
   void stop() override {
     text_token_generator_->stop();
   }

extension/llm/runner/text_llm_runner.h

@@ -18,6 +18,7 @@
 #include <unordered_map>
 
 #include <executorch/extension/llm/runner/irunner.h>
+#include <executorch/extension/llm/runner/multimodal_input.h>
 #include <executorch/extension/llm/runner/stats.h>
 #include <executorch/extension/llm/runner/text_decoder_runner.h>
 #include <executorch/extension/llm/runner/text_prefiller.h>
@@ -101,8 +102,6 @@ class ET_EXPERIMENTAL TextLLMRunner : public IRunner {
       std::function<void(const std::string&)> token_callback = {},
       std::function<void(const Stats&)> stats_callback = {}) override;
 
-  using IRunner::prefill; // Bring in string convenience overload
-
   /**
    * Prefill multimodal inputs into the KV cache without generating.
    * Only text inputs are processed; non-text inputs are skipped.
@@ -116,6 +115,16 @@ class ET_EXPERIMENTAL TextLLMRunner : public IRunner {
       int32_t num_bos = 0,
       int32_t num_eos = 0) override;
 
+  /**
+   * Convenience overload: prefill a single text prompt.
+   */
+  ::executorch::runtime::Error
+  prefill(const std::string& prompt, int32_t num_bos = 0, int32_t num_eos = 0) {
+    std::vector<MultimodalInput> inputs;
+    inputs.emplace_back(MultimodalInput(prompt));
+    return prefill(inputs, num_bos, num_eos);
+  }
+
   /**
    * Prefill a text prompt using GenerationConfig.
    * Deprecated: prefer prefill(prompt, num_bos, num_eos).