fix: store actual rendered prompt artifact instead of contextpack.toString()

Extract message-rendering logic from LlamaCppInferenceProvider into a
provider-agnostic PromptRenderer in core:inference. SessionOrchestrator
now serializes the rendered ChatMessage list as JSON for the prompt
artifact, so the audit trail reflects what the LLM actually saw.

Resolves the TODO at SessionOrchestrator.kt:691.

core/inference/src/main/kotlin/com/correx/core/inference/PromptRenderer.kt

@@ -0,0 +1,41 @@
+package com.correx.core.inference
+
+import com.correx.core.context.model.ContextLayer
+import com.correx.core.context.model.ContextPack
+import com.correx.core.context.model.EntryRole
+import kotlinx.serialization.Serializable
+
+@Serializable
+data class ChatMessage(
+    val role: String,
+    val content: String,
+)
+
+object PromptRenderer {
+    fun render(contextPack: ContextPack): List<ChatMessage> {
+        val sorted = contextPack.layers.entries.sortedBy { it.key.ordinal }
+        val systemContent = sorted
+            .filter { it.key == ContextLayer.L0 }
+            .flatMap { it.value }
+            .joinToString("\n\n") { it.content }
+            .takeIf { it.isNotBlank() }
+        val conversationMessages = sorted.filter { it.key != ContextLayer.L0 }.flatMap { (_, entries) ->
+            entries.map { entry ->
+                ChatMessage(
+                    role = when (entry.role) {
+                        EntryRole.SYSTEM -> "system"
+                        EntryRole.ASSISTANT -> "assistant"
+                        EntryRole.TOOL -> "tool"
+                        EntryRole.USER -> "user"
+                    },
+                    content = entry.content,
+                )
+            }
+        }
+        val messages = buildList {
+            systemContent?.let { add(ChatMessage("system", it)) }
+            addAll(conversationMessages)
+        }
+        return messages.ifEmpty { listOf(ChatMessage("user", "")) }
+    }
+}

core/kernel/src/main/kotlin/com/correx/core/kernel/orchestration/SessionOrchestrator.kt

@@ -61,6 +61,7 @@ import com.correx.core.inference.InferenceRepository
 import com.correx.core.inference.InferenceRequest
 import com.correx.core.inference.InferenceResponse
 import com.correx.core.inference.InferenceRouter
+import com.correx.core.inference.PromptRenderer
 import com.correx.core.inference.ResponseFormat
 import com.correx.core.inference.Tokenizer
 import com.correx.core.inference.ToolCallRequest
@@ -91,6 +92,7 @@ import kotlinx.coroutines.TimeoutCancellationException
 import kotlinx.coroutines.withContext
 import kotlinx.coroutines.withTimeout
 import kotlinx.datetime.Clock
+import kotlinx.serialization.encodeToString
 import kotlinx.serialization.json.Json
 import kotlinx.serialization.json.JsonObject
 import org.slf4j.LoggerFactory
@@ -688,8 +690,8 @@ abstract class SessionOrchestrator(
                 ),
         )
 
-        // TODO: capture rendered prompt; currently contextpack.toString()
-        val promptArtifactId = artifactStore.put(contextPack.toString().toByteArray())
+        val renderedMessages = PromptRenderer.render(contextPack)
+        val promptArtifactId = artifactStore.put(Json.encodeToString(renderedMessages).toByteArray())
         emit(sessionId, InferenceStartedEvent(requestId, sessionId, stageId, provider.id, promptArtifactId))
 
         if (isCancelled(sessionId)) return InferenceResult.Cancelled