feat: pydantic use oi helper functions

[GeLi2001]

Note

Refactors semantic_conventions to use OpenInference helper functions/constants for emitting span, message, tool, and token count attributes, updating keys and serialization accordingly.

• Refactor: replace direct SpanAttributes/MessageAttributes/Tool usage with OI helpers*
  • Use get_span_kind_attributes, get_llm_system_attributes, get_llm_model_name_attributes, get_llm_token_count_attributes, get_session_attributes, get_input_attributes, get_output_attributes, get_llm_invocation_parameter_attributes, and get_tool_attributes.
  • Swap attribute keys to _attributes constants (e.g., LLM_INPUT_MESSAGES, LLM_OUTPUT_MESSAGES, MESSAGE_ROLE, MESSAGE_CONTENT, MESSAGE_TOOL_CALLS, TOOL_CALL_*, TOOL_NAME, TOOL_DESCRIPTION).
  • Compute token counts via TokenCount dict (prompt/completion/total) and skip for agent spans.
• Messages and events
  • Update message extraction/flattening to use new MESSAGE_* and LLM_* keys; use helpers for input/output values.
• Tools
  • Emit tool span attributes via get_tool_attributes (yielding only TOOL_NAME/TOOL_DESCRIPTION).
  • Update tool call attributes to TOOL_CALL_* and tool definitions to LLM_TOOLS.<idx>.* with _json_serialize for TOOL_JSON_SCHEMA.

^{Written by Cursor Bugbot for commit debbdf5. This will update automatically on new commits. Configure here.}