* fix: prevent duplicate spend logs in Responses API for non-OpenAI providers
Fixes#15740
This fixes a logging duplication bug where using kwargs.pop() removed
the litellm_logging_obj before passing kwargs to internal acompletion()
calls, causing duplicate spend log entries for providers without native
Responses API support (Anthropic, Gemini, etc).
By changing from pop() to get(), the logging object is preserved and
reused across the internal completion call, preventing duplicate entries
and maintaining correct cost tracking.
* test: add test for logging object preservation in responses API
Verify that litellm_logging_obj is preserved in kwargs when calling
responses(), ensuring no duplicate spend log entries are created.