Fix Langfuse callback skipping for Anthropic streaming passthrough

Fixes #17476

When using Anthropic pass-through endpoints with streaming and Langfuse
callbacks, streaming requests were completing successfully but failing to
log to Langfuse. Non-streaming requests worked correctly.

Root cause:
- _handle_logging_anthropic_collected_chunks() successfully builds the
  complete response from chunks
- However, kwargs only included response_cost and model, missing the
  complete_streaming_response key
- In litellm_logging.py, Langfuse callback checks for
  complete_streaming_response and skips logging when None

Solution:
- Add complete_streaming_response to kwargs dict before returning
- This ensures callbacks like Langfuse can access the complete response
  and properly log streaming requests

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

litellm/proxy/pass_through_endpoints/llm_provider_handlers/anthropic_passthrough_logging_handler.py

@@ -218,6 +218,9 @@ class AnthropicPassthroughLoggingHandler:
             logging_obj=litellm_logging_obj,
         )
 
+        # Add complete_streaming_response to kwargs so callbacks like Langfuse can access it
+        kwargs["complete_streaming_response"] = complete_streaming_response
+
         return {
             "result": complete_streaming_response,
             "kwargs": kwargs,