use litellm sdk

[xzrderek]

Note

Moves the gateway from HTTP proxying to direct LiteLLM SDK usage with Langfuse OTEL, simplifying deployment and unifying tracing.

• Proxy runtime: litellm.py now calls litellm.acompletion directly (supports SSE streaming), injects api_key, builds Langfuse metadata/tags (incl. UUID7 insertion_id), maps user_id→trace_user_id, and registers insertions in Redis after success
• App wiring: On startup, sets LANGFUSE_* env and litellm.callbacks=["langfuse_otel"]; removes catch‑all proxy route and drops litellm_url from ProxyConfig
• Container/compose: Dockerfile builds package locally and runs only the gateway; docker‑compose removes litellm-backend service and network; config updates to callbacks: ["langfuse_otel"]
• Docs: README updated to reflect SDK‑direct flow, new env defaults (LANGFUSE_HOST), and simplified setup
• Deps: Bump litellm (>=1.81,<1.82), add OTEL packages for proxy extra, update lockfile; remove langchain-fireworks extra
• Tracing adapter: Generalize OTEL attribute parsing and provider‑agnostic *.choices extraction to set finish_reason

^{Written by Cursor Bugbot for commit 4a3cc04. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.