Port main branch updates to refactored api/common/conserver structure

[pavanputhra]

Summary

Backports 7 commits from main (up to 5f3350c) into the optimize-docker-images split layout, adapting all file paths and imports from the old server/ structure.

• New common/lib/openai_client.py — shared get_openai_client() / get_async_openai_client() supporting OpenAI, Azure OpenAI, and LiteLLM proxy; all links/storage now call this instead of constructing clients inline
• deepgram_link — add LiteLLM proxy transcription path (transcribe_via_litellm), fix fd leak in audio temp file handling, make confidence check optional (not available on LiteLLM path)
• wtf_transcribe — updated for new vfun /wtf API: simplified create_wtf_analysis (pass response body directly), file-binary field name, language option, diarize default→False, min-duration default→0, accept status 200 only
• api — /config endpoint uses Configuration.get_config() instead of reading the YAML file directly
• tests — add mock_get_client patches to analyze_and_label and detect_engagement tests; fix test_external_ingress to patch api.index_vcon instead of api.index_vcon_parties
• docs — add Langfuse integration and OTel Collector fan-out documentation to monitoring.md
• .gitignore — add litellm_config.yaml (local dev file with credentials, was untracked)

Test plan

• Run existing test suite via docker compose run --rm conserver poetry run pytest
• Verify /config endpoint returns config correctly
• Verify deepgram transcription works with both direct Deepgram key and LiteLLM proxy
• Verify wtf_transcribe works against updated vfun /wtf endpoint

🤖 Generated with Claude Code