Fixes#3982
This commit adds tool call event emission to all provider-specific LLM
streaming implementations. Previously, only text chunks were emitted
during streaming, but tool call information was missing.
Changes:
- Update BaseLLM._emit_stream_chunk_event to infer call_type from
tool_call presence when not explicitly provided
- Add tool call event emission in OpenAI provider streaming
- Add tool call event emission in Azure provider streaming
- Add tool call event emission in Gemini provider streaming
- Add tool call event emission in Bedrock provider streaming
- Add tool call event emission in Anthropic provider streaming
- Add comprehensive tests for tool call streaming events
The fix ensures that LLMStreamChunkEvent is emitted with:
- call_type=LLMCallType.TOOL_CALL when tool calls are received
- tool_call dict containing id, function (name, arguments), type, index
- chunk containing the tool call arguments being streamed
Co-Authored-By: João <joao@crewai.com>
b70c4499d7