feat(pydentic_ai): tools & multiple input/output messages support

[satyadevai]

Closes #2476 #2475

Multiple Tool Output Messages from LLM

Tools with Tool Details & Output:

Multiple Input Messages - Multiple Tool Calls & Tools Output

---

Note

Extends instrumentation to capture multi-part input/output messages and tool definitions/args/responses, sets span status codes, and adds a multi-tool agent example with updated tests.

• Instrumentation (semantic conventions):
  • Support multi-part message contents via MessageContentAttributes and indexed MESSAGE_CONTENTS for gen_ai.input/output.messages.
  • Capture tool metadata: LLM_TOOLS from both output_tools and function_tools with JSON schema; include tool_arguments → TOOL_PARAMETERS and tool_response → OUTPUT_VALUE.
  • Normalize tool call fields: index tool calls in messages; JSON-serialize non-string arguments; JSON-serialize final_result output.
  • Add constants: PydanticTools.TOOL_ARGUMENTS/TOOL_RESPONSE, PydanticModelRequestParameters.FUNCTION_TOOLS, and adjust tool PARAMETERS to properties.
• Span processor:
  • Set span StatusCode.OK on success; mark StatusCode.ERROR with exception details on failures.
• Examples:
  • New examples/multi_tool_agent.py demonstrating a multi-tool Agent with OTLP export and two tools (get_weather, get_time).
• Tests:
  • Update assertions to handle v1 vs v2 message formats and verify tool call args, tools metadata, and token counts.

^{Written by Cursor Bugbot for commit 284e0f0. This will update automatically on new commits. Configure here.}