feat(llm): add LLM call support to worker IPC chain

- ProviderManager: align API with contracts ProviderAdapter
- WorkerProtocol: add llm.request/llm.response message types
- WorkerRuntime: add call_llm() for worker→parent→LLM flow
- WorkerManager: support tool_registry and provider_manager injection

P1-1 complete, P1-2 protocol layer complete.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

packages/llm/src/ProviderManager.ts

@@ -7,13 +7,11 @@
  * @module packages/llm/src/ProviderManager
  */
 
-import type { ProviderAdapter } from '@aircoding/contracts'
+import type { ProviderAdapter, ProviderCompletionInput, ProviderStreamEvent, ProviderCapabilityMatrix, ModelID, ProviderID } from '@aircoding/contracts'
 
-// Local type definitions (contract types not yet finalized)
-type CompleteOptions = { max_tokens?: number; temperature?: number; top_p?: number; system?: string }
 type StreamEvent = { type: 'text' | 'thinking' | 'done'; content?: string; reason?: string }
 type ModelRequirement = { model: string; provider?: string; min_output_tokens?: number; prefers_thinking?: boolean; requires_tools?: boolean }
-type ModelAssignment = { provider: string; model: string; adapter: any; capabilities?: any }
+type ModelAssignment = { provider: string; model: string; adapter: ProviderAdapter; capabilities?: ReturnType<CapabilityMatrixRegistry['lookup']> }
 
 import { ModelConfigLoader, createModelConfigLoader } from './ModelConfigLoader.js'
 import { CapabilityMatrixRegistry, createCapabilityMatrixRegistry } from './CapabilityMatrix.js'
@@ -80,34 +78,118 @@ export class ProviderManager {
 
   /**
    * Complete a request with the current model.
+   * Returns AsyncIterable of ProviderStreamEvent per contracts §15.
    */
-  async complete(
+  async *complete(
+    input: ProviderCompletionInput
+  ): AsyncGenerator<ProviderStreamEvent> {
+    const adapter = this.current_adapter || this.adapters.get(input.provider_id || 'anthropic')
+    if (!adapter) {
+      throw new Error('No adapter selected. Call select_model first.')
+    }
+
+    yield* adapter.complete(input)
+  }
+
+  /**
+   * Complete and collect all events into a single response.
+   */
+  async complete_text(
+    messages: unknown[],
+    options: { model?: string; max_tokens?: number; temperature?: number; system?: string } = {}
+  ): Promise<{ content: string; usage?: { input_tokens: number; output_tokens: number } }> {
+    const adapter = this.current_adapter
+    if (!adapter) {
+      throw new Error('No adapter selected. Call select_model first.')
+    }
+
+    const model_id = options.model || 'claude-haiku-4-5-20251001'
+    const input: ProviderCompletionInput = {
+      provider_id: 'anthropic',
+      model_id: model_id as ModelID,
+      canonical_format: 'anthropic',
+      messages,
+      max_output_tokens: options.max_tokens || 4096,
+      temperature: options.temperature,
+      system: options.system
+    }
+
+    let content = ''
+    let usage: { input_tokens: number; output_tokens: number } | undefined
+
+    for await (const event of adapter.complete(input)) {
+      if (event.type === 'content_delta') {
+        const payload = event.payload as { type: string; text?: string; thinking?: string }
+        if (payload.type === 'text_delta') {
+          content += payload.text || ''
+        } else if (payload.type === 'thinking_delta') {
+          // Accumulate thinking for reference but don't include in content
+        }
+      } else if (event.type === 'message_stop') {
+        const payload = event.payload as { usage?: { output_tokens: number } }
+        if (payload.usage) {
+          usage = { input_tokens: 0, output_tokens: payload.usage.output_tokens }
+        }
+      }
+    }
+
+    return { content, usage }
+  }
+
+  /**
+   * Stream a completion request (passthrough to adapter).
+   */
+  async *stream_complete(
+    input: ProviderCompletionInput
+  ): AsyncGenerator<ProviderStreamEvent> {
+    const adapter = this.current_adapter
+    if (!adapter) {
+      throw new Error('No adapter selected. Call select_model first.')
+    }
+
+    yield* adapter.complete(input)
+  }
+
+  /**
+   * Legacy overload: select model then complete.
+   */
+  async complete_after_select(
     messages: unknown[],
     assignment: ModelAssignment,
-    options: CompleteOptions = {}
+    options: { max_tokens?: number; temperature?: number } = {}
   ): Promise<{ content: string; usage?: { input_tokens: number; output_tokens: number } }> {
     const adapter = assignment.adapter || this.current_adapter
     if (!adapter) {
       throw new Error('No adapter selected. Call select_model first.')
     }
 
-    return adapter.complete(messages as any, { model: assignment.model }, options)
-  }
-
-  /**
-   * Stream a completion request.
-   */
-  async *stream_complete(
-    messages: unknown[],
-    assignment: ModelAssignment,
-    options: CompleteOptions = {}
-  ): AsyncGenerator<StreamEvent> {
-    const adapter = assignment.adapter || this.current_adapter
-    if (!adapter) {
-      throw new Error('No adapter selected. Call select_model first.')
+    const input: ProviderCompletionInput = {
+      provider_id: assignment.provider as ProviderID || 'anthropic',
+      model_id: assignment.model as ModelID,
+      canonical_format: 'anthropic',
+      messages,
+      max_output_tokens: options.max_tokens || 4096,
+      temperature: options.temperature
     }
 
-    yield* adapter.stream_complete(messages as any, { model: assignment.model }, options)
+    let content = ''
+    let usage: { input_tokens: number; output_tokens: number } | undefined
+
+    for await (const event of adapter.complete(input)) {
+      if (event.type === 'content_delta') {
+        const payload = event.payload as { type: string; text?: string }
+        if (payload.type === 'text_delta') {
+          content += payload.text || ''
+        }
+      } else if (event.type === 'message_stop') {
+        const payload = event.payload as { usage?: { output_tokens: number } }
+        if (payload.usage) {
+          usage = { input_tokens: 0, output_tokens: payload.usage.output_tokens }
+        }
+      }
+    }
+
+    return { content, usage }
   }
 
   /**