feat: add agent workflow metrics collection

Add comprehensive OpenTelemetry-based metrics for agent observability:

- Workflow completion/failure tracking with duration measurements
- Step execution counters for performance monitoring
- Tool usage tracking with normalized tool names
- Non-blocking telemetry emission with named async tasks
- Comprehensive unit and integration test coverage
- Graceful handling when telemetry is disabled

llama_stack/providers/inline/agents/meta_reference/agents.py

@@ -38,6 +38,7 @@ from llama_stack.apis.inference import (
     UserMessage,
 )
 from llama_stack.apis.safety import Safety
+from llama_stack.apis.telemetry import Telemetry
 from llama_stack.apis.tools import ToolGroups, ToolRuntime
 from llama_stack.apis.vector_io import VectorIO
 from llama_stack.core.datatypes import AccessRule
@@ -64,6 +65,7 @@ class MetaReferenceAgentsImpl(Agents):
         tool_runtime_api: ToolRuntime,
         tool_groups_api: ToolGroups,
         policy: list[AccessRule],
+        telemetry_api: Telemetry | None = None,
     ):
         self.config = config
         self.inference_api = inference_api
@@ -71,6 +73,7 @@ class MetaReferenceAgentsImpl(Agents):
         self.safety_api = safety_api
         self.tool_runtime_api = tool_runtime_api
         self.tool_groups_api = tool_groups_api
+        self.telemetry_api = telemetry_api
 
         self.in_memory_store = InmemoryKVStoreImpl()
         self.openai_responses_impl: OpenAIResponsesImpl | None = None
@@ -130,6 +133,7 @@ class MetaReferenceAgentsImpl(Agents):
             vector_io_api=self.vector_io_api,
             tool_runtime_api=self.tool_runtime_api,
             tool_groups_api=self.tool_groups_api,
+            telemetry_api=self.telemetry_api,
             persistence_store=(
                 self.persistence_store if agent_info.enable_session_persistence else self.in_memory_store
             ),