feat: implement CreateGrant handler and grant-aware approval engine

- GlobalStreamHandler.handleCreateGrant validates scope (SESSION/STAGE)
  and sends ProtocolError to client on validation failure instead of
  silent log.warn + drop.
- Derive event stageId directly from GrantScope type, removing
  redundant stageIdForEvent tuple.
- SessionOrchestrator integrates ApprovalEngine to check active grants
  before requesting user approval for T2+ tool calls.
- ContextEntry gains EntryRole (SYSTEM/USER/ASSISTANT/TOOL) field for
  proper chat message role mapping.
- RouterContextBuilder.build is now suspend; uses Tokenizer for
  accurate token estimation with fallback to character-based estimate.
- LlamaCppInferenceProvider maps EntryRole to ChatMessage role instead
  of heuristic layer-based inference.

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

@@ -2,6 +2,7 @@ package com.correx.infrastructure.inference.llama.cpp
 
 import com.correx.core.context.model.ContextLayer
 import com.correx.core.context.model.ContextPack
+import com.correx.core.context.model.EntryRole
 import com.correx.core.events.types.ProviderId
 import com.correx.core.inference.CapabilityScore
 import com.correx.core.inference.FinishReason
@@ -149,17 +150,17 @@ class LlamaCppInferenceProvider(
             .flatMap { it.value }
             .joinToString("\n\n") { it.content }
             .takeIf { it.isNotBlank() }
-        val conversationMessages = sorted.filter { it.key != ContextLayer.L0 }.flatMap { (layer, entries) ->
-            when (layer) {
-                ContextLayer.L1 -> entries.map { ChatMessage("user", it.content) }
-                ContextLayer.L2 -> entries.map {
-                    ChatMessage(
-                        role = if (it.sourceType == "assistant") "assistant" else "user",
-                        content = it.content,
-                    )
-                }
-
-                else -> emptyList()
+        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 {
@@ -168,4 +169,4 @@ class LlamaCppInferenceProvider(
         }
         return messages.ifEmpty { listOf(ChatMessage("user", "")) }
     }
-}
+}

infrastructure/src/main/kotlin/com/correx/infrastructure/InfrastructureModule.kt

@@ -10,6 +10,7 @@ import com.correx.core.events.EventDispatcher
 import com.correx.core.events.stores.EventStore
 import com.correx.core.inference.InferenceProvider
 import com.correx.core.inference.InferenceRouter
+import com.correx.core.inference.Tokenizer
 import com.correx.core.router.DefaultRouterContextBuilder
 import com.correx.core.router.DefaultRouterFacade
 import com.correx.core.router.DefaultRouterReducer
@@ -120,12 +121,13 @@ object InfrastructureModule {
         eventStore: EventStore,
         inferenceRouter: InferenceRouter,
         config: RouterConfig = RouterConfig(),
+        tokenizer: Tokenizer? = null,
     ): RouterFacade {
         val reducer = DefaultRouterReducer()
         val projector = RouterProjector(reducer)
         val replayer = DefaultEventReplayer(eventStore, projector)
         val repository = DefaultRouterRepository(replayer)
-        val contextBuilder = DefaultRouterContextBuilder(config)
+        val contextBuilder = DefaultRouterContextBuilder(config, tokenizer)
         return DefaultRouterFacade(repository, contextBuilder, inferenceRouter, eventStore, config)
     }
 }