This fixes GitHub issue #4170 where token usage metrics were not being
updated when using litellm with streaming responses and async calls.
Changes:
- Add token usage tracking to _handle_non_streaming_response
- Add token usage tracking to _ahandle_non_streaming_response
- Add token usage tracking to _ahandle_streaming_response
- Fix sync streaming to track usage in both code paths
- Convert usage objects to dicts before passing to _track_token_usage_internal
- Add comprehensive tests for token usage tracking in all scenarios
Co-Authored-By: João <joao@crewai.com>
