feat(ai-core): wire function calling into the LLM pipeline

- llm: ChatStreamWithTools sends OpenAI-format tools; streamed tool_call
  deltas are reassembled by index into complete ToolCalls on the Done token
- model: LLMMessage gains ToolCallID/ToolCalls; StreamToken gains ToolCalls
- orchestrator: new ToolExecutor interface and streamWithTools loop —
  execute requested tools, feed results back, up to 3 rounds, final round
  forces a plain text answer; normal chats still stream token by token.
  Also adds Pipeline.FinishRound (assistant reply recording + post-round
  hook), wired to transports in the next commit.
- main: inject the plugin registry via toolExecAdapter

Includes an httptest-based test covering delta reassembly, tool
execution, and result feedback.

backend/cmd/cyrene/main.go

@@ -193,6 +193,7 @@ func main() {
 		memExtractor := memory.NewExtractor(memSvc, &memory.LLMAdapterBridge{Adapter: memAdapter})
 		orch.SetMemoryExtractor(memExtractor)
 	}
+	orch.SetToolExecutor(&toolExecAdapter{reg: reg})
 	logger.Infof("[orchestrator] pipeline ready")
 
 	// === 11.5 Platform Bridge ===
@@ -330,6 +331,22 @@ func (a *perfAdapter) History() []map[string]any {
 	return result
 }
 
+// toolExecAdapter adapts the plugin registry to orchestrator.ToolExecutor.
+type toolExecAdapter struct{ reg *plugin.Registry }
+
+func (t *toolExecAdapter) Tools() []llm.ToolSpec {
+	defs := t.reg.GetAllToolDefinitions()
+	specs := make([]llm.ToolSpec, len(defs))
+	for i, d := range defs {
+		specs[i] = llm.ToolSpec{Name: d.Name, Description: d.Description, Parameters: d.Parameters}
+	}
+	return specs
+}
+
+func (t *toolExecAdapter) Execute(ctx context.Context, name string, params map[string]any) (string, error) {
+	return t.reg.ExecuteTool(ctx, name, params)
+}
+
 // bridgeConfigAdapter adapts bridge.ConfigStore to api.BridgeConfigManager.
 type bridgeConfigAdapter struct{ *bridge.ConfigStore }
 func (a *bridgeConfigAdapter) GetAdapters() []map[string]any {