Fix LiteLLM cost tracking for provider-prefixed models

[ShreySatapara]

Summary

Fixes missing cost tracking in Conversation.run() when using provider-prefixed model names (e.g. vertex_ai/claude-sonnet-4-5@..., azure/gpt-5.2-chat).
Previously, telemetry’s LiteLLM cost fallback stripped the provider and passed only the bare model name to litellm_completion_cost(), causing LiteLLM to warn “LLM Provider NOT provided” and return None for cost. This PR preserves the provider by passing custom_llm_provider=<provider> and model=<bare_model> to LiteLLM’s cost calculator.

Also adds regression tests to ensure provider and model are propagated correctly for cost calculation.

[fill in a summary of this PR]

Checklist

• [✔ ] If the PR is changing/adding functionality, are there tests to reflect this?
• • Added unit tests in tests/sdk/test_telemetry.py that assert litellm_completion_cost() is called with custom_llm_provider and the correct bare model (covers Vertex AI + Azure).
• [ ✔] If there is an example, have you run the example to make sure that it works?
• • Conducted Experiment before and after doing the suggested changes, cost is getting reflected in telemetry logs for model names with provided which was showing $0.00 before making suggested changes.
• [ ✔] If there are instructions on how to run the code, have you followed the instructions and made sure that it works?
• • Ran local checks per DEVELOPMENT.md: make format, make lint, uv run pre-commit run --all-files, and uv run pytest.
• If the feature is significant enough to require documentation, is there a PR open on the OpenHands/docs repository with the same branch name?
• • N/A (bugfix; no user-facing docs update required).
• Is the github CI passing?
• • I did the pre-commit checks and all the offline checks, not sure which additional workflow I need to execute.

[enyst]

Thank you! Litellm can be unexpected sometimes, LLMs do have "provider/model" names so it's doing the split itself in the regular completion method... but maybe not in completion_cost? I set off OpenHands agent to take a look too 😅

[enyst]

Thank you for the fix!