feat(inference): capture reasoning_content on responses (WIP, provider side)

Adds a nullable `reasoning` field to InferenceResponse and maps the
llama.cpp / OpenAI-compat `reasoning_content` field into it, so local models
that emit their chain-of-thought have it captured. Provider/model side only —
the emit-site event field (reasoningArtifactId) + CAS storage in the
orchestrator are still pending (tracked in Vikunja Correx #4).

infrastructure/inference/llama_cpp/src/main/kotlin/com/correx/infrastructure/inference/llama/cpp/LlamaCppApiModels.kt

@@ -24,6 +24,7 @@ data class ChatCompletionRequest(
 data class LlamaCppChatMessage(
     val role: String,
     val content: String?,
+    @SerialName("reasoning_content") val reasoningContent: String? = null,
     @SerialName("tool_calls") val toolCalls: List<ToolCallRequest> = emptyList(),
     @SerialName("tool_call_id") val toolCallId: String? = null,
 )

infrastructure/inference/llama_cpp/src/main/kotlin/com/correx/infrastructure/inference/llama/cpp/LlamaCppInferenceProvider.kt

@@ -215,6 +215,7 @@ class LlamaCppInferenceProvider(
             ),
             latencyMs = System.currentTimeMillis() - startTime,
             toolCalls = toolCalls,
+            reasoning = message.reasoningContent?.takeIf { it.isNotBlank() },
         )
     }
 

infrastructure/inference/openai_compat/src/main/kotlin/com/correx/infrastructure/inference/openai/OpenAiApiModels.kt

@@ -28,6 +28,7 @@ data class OpenAiChatCompletionRequest(
 data class OpenAiChatMessage(
     val role: String,
     val content: String? = null,
+    @SerialName("reasoning_content") val reasoningContent: String? = null,
     @SerialName("tool_calls") val toolCalls: List<ToolCallRequest> = emptyList(),
     @SerialName("tool_call_id") val toolCallId: String? = null,
 )