Add support for OTel GenAI Semantic Convention patterns in LLO handler (#460)

*Description of changes:*
Handle specific span events by recognizing them via event.name. Move
their associated attributes into Log events (as content) and remove them
from the original Span Event.

* gen_ai.assistant.message (output)
* gen_ai.tool.message (input)
* gen_ai.user.message (input)
* gen_ai.system.message (input)
* gen_ai.choice (output)

### Test
```
{
    "resource": {
        "attributes": {
            "aws.local.service": "langgraph-min",
            "aws.service.type": "gen_ai_agent",
            "telemetry.sdk.language": "python",
            "service.name": "langgraph-min",
            "cloud.resource_id": "fastapi-app",
            "aws.log.group.names": "test_trace-min-4",
            "telemetry.sdk.version": "1.33.1",
            "telemetry.auto.version": "0.11.0.dev0-aws",
            "telemetry.sdk.name": "opentelemetry"
        }
    },
    "scope": {
        "name": "strands.telemetry.tracer"
    },
    "timeUnixNano": 1756004845660372000,
    "observedTimeUnixNano": 1756004845662834000,
    "severityNumber": 9,
    "severityText": "",
    "body": {
        "output": {
            "messages": [
                {
                    "content": {
                        "message": "# Weather Forecast for New York City 🌤️\n\nHere's the current weather forecast for New York City (Manhattan area):\n\n## **Current Conditions (Tonight - Saturday Night, Aug 23)**\n🌙 **Tonight**: Partly Cloudy  \n- **Temperature**: Low around 70°F  \n- **Wind**: South 7-13 mph  \n- **Precipitation**: 0% chance  \n\n## **This Weekend**\n\n### **Sunday (Aug 24)**\n🌤️ **Partly Sunny**  \n- **High**: 77°F  \n- **Wind**: South 7-16 mph  \n- **Precipitation**: 10% chance  \n\n### **Sunday Night**  \n🌧️ **Chance Rain Showers** (after 8pm)  \n- **Low**: 72°F  \n- **Wind**: South 6-15 mph  \n- **Precipitation**: 40% chance  \n\n### **Monday (Aug 25)**  \n🌦️ **Chance Rain Showers**  \n- **High**: 80°F  \n- **Wind**: Southwest 6-10 mph  \n- **Precipitation**: 40% chance  \n\n## **Looking Ahead (Tuesday - Saturday)**\n\n**Tuesday**: ☀️ Sunny, High 76°F - Beautiful day!  \n**Wednesday**: 🌤️ Mostly Sunny, High 73°F  \n**Thursday**: 🌤️ Mostly Sunny, High 76°F  \n**Friday**: 🌤️ Mostly Sunny, High 76°F  \n**Saturday**: 🌤️ Mostly Sunny, High 75°F  \n\n## **Key Highlights**\n- **Pleasant temperatures** in the mid-70s to near 80°F\n- **Possible rain** Sunday night through Monday morning\n- **Great weather** returning Tuesday with sunny skies\n- **Comfortable humidity** levels expected\n- **Light to moderate winds** throughout the period\n\nThis forecast shows typical late August weather for NYC with comfortable temperatures and mostly pleasant conditions, with just a brief chance of showers early in the week before clearing up nicely!\n",
                        "finish_reason": "end_turn"
                    },
                    "role": "assistant"
                }
            ]
        },
        "input": {
            "messages": [
                {
                    "content": "You are a weather assistant with HTTP capabilities. You can:\n\n1. Make HTTP requests to the National Weather Service API\n2. Process and display weather forecast data\n3. Provide weather information for locations in the United States\n\nWhen retrieving weather information:\n1. First get the coordinates or grid information using https://api.weather.gov/points/{latitude},{longitude} or https://api.weather.gov/points/{zipcode}\n2. Then use the returned forecast URL to get the actual forecast\n\nWhen displaying responses:\n- Format weather data in a human-readable way\n- Highlight important information like temperature, precipitation, and alerts\n- Handle errors appropriately\n- Convert technical terms to user-friendly language\n\nAlways explain the weather conditions clearly and provide context for the forecast.\n",
                    "role": "system"
                },
                {
                    "content": {
                        "content": "[{\"text\": \"What is the weather in New York?\"}]"
                    },
                    "role": "user"
                }
            ]
        }
    },
    "attributes": {
        "event.name": "strands.telemetry.tracer"
    },
    "flags": 1,
    "traceId": "68aa81cd356fb8ea277cca6fbd34640d",
    "spanId": "5a3802fa510a2725"
}
```
By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.

Co-authored-by: Michael He <[email protected]>

Author: mxiamxia

aws-opentelemetry-distro/src/amazon/opentelemetry/distro/llo_handler.py

@@ -16,6 +16,7 @@
 ROLE_SYSTEM = "system"
 ROLE_USER = "user"
 ROLE_ASSISTANT = "assistant"
+ROLE_TOOL = "tool"
 
 _logger = logging.getLogger(__name__)
 
@@ -137,6 +138,35 @@ class PatternConfig(TypedDict, total=False):
         "role": ROLE_USER,
         "source": "prompt",
     },
+    # OTel GenAI Semantic Convention used by the latest Strands SDK
+    # References:
+    # - OTel GenAI SemConv: https://opentelemetry.io/docs/specs/semconv/gen-ai/gen-ai-events/
+    # - Strands SDK PR(introduced in v0.1.9): https://github.com/strands-agents/sdk-python/pull/319
+    "gen_ai.user.message": {
+        "type": PatternType.DIRECT,
+        "role": ROLE_USER,
+        "source": "prompt",
+    },
+    "gen_ai.assistant.message": {
+        "type": PatternType.DIRECT,
+        "role": ROLE_ASSISTANT,
+        "source": "output",
+    },
+    "gen_ai.system.message": {
+        "type": PatternType.DIRECT,
+        "role": ROLE_SYSTEM,
+        "source": "prompt",
+    },
+    "gen_ai.tool.message": {
+        "type": PatternType.DIRECT,
+        "role": ROLE_TOOL,
+        "source": "prompt",
+    },
+    "gen_ai.choice": {
+        "type": PatternType.DIRECT,
+        "role": ROLE_ASSISTANT,
+        "source": "output",
+    },
 }
 
 
@@ -214,6 +244,7 @@ def _collect_all_llo_messages(self, span: ReadableSpan, attributes: types.Attrib
         for attr_key, value in attributes.items():
             if attr_key in self._exact_match_patterns:
                 config = self._pattern_configs[attr_key]
+
                 messages.append(
                     {"content": value, "role": config.get("role", "unknown"), "source": config.get("source", "unknown")}
                 )
@@ -279,6 +310,12 @@ def _collect_llo_attributes_from_span(self, span: ReadableSpan) -> Dict[str, Any
         # Collect from span events
         if span.events:
             for event in span.events:
+                # Check if event name itself is an LLO pattern (e.g., "gen_ai.user.message")
+                if self._is_llo_attribute(event.name):
+                    # Put all event attributes as the content as LLO in log event
+                    all_llo_attributes[event.name] = dict(event.attributes) if event.attributes else {}
+
+                # Also check traditional pattern - LLO attributes within event attributes
                 if event.attributes:
                     for key, value in event.attributes.items():
                         if self._is_llo_attribute(key):
@@ -372,6 +409,10 @@ def _filter_span_events(self, span: ReadableSpan) -> None:
         updated_events = []
 
         for event in span.events:
+            # Skip entire event if event name is an LLO pattern
+            if self._is_llo_attribute(event.name):
+                continue
+
             if not event.attributes:
                 updated_events.append(event)
                 continue
@@ -417,7 +458,7 @@ def _group_messages_by_type(self, messages: List[Dict[str, Any]]) -> Dict[str, L
             elif role == ROLE_ASSISTANT:
                 output_messages.append(formatted_message)
             else:
-                # Route based on source for non-standard roles
+                # Route based on source for non-standard roles including tool
                 if any(key in message.get("source", "") for key in ["completion", "output", "result"]):
                     output_messages.append(formatted_message)
                 else: