Add F1 radio RAG demo (full Opik eval loop)

[fschlz]

What

A new use-cases/f1_radio_rag/ demo: a small, runnable Typer CLI that walks the entire Opik evaluation-and-improvement loop for a RAG use case — summarising F1 team-radio messages across a race weekend.

The loop (one CLI command each)

• ingest — load radio messages into a local ChromaDB store (offline)
• ask — retrieve + summarise with Claude (via litellm), traced in Opik
• eval — build an Opik dataset + test suite; run plain-English assertions (run_tests) and the ContextRecall + Hallucination metrics (evaluate)
• optimize — MetaPromptOptimizer improves the summariser prompt against the dataset
• promote — save the optimised prompt to the Prompt Library (versioned)
• run-all — the whole chain

Notes

• uv project (pyproject.toml + uv.lock); pip install line also in the README.
• Every command has a DRY_RUN path — runs with no credentials and prints useful output.
• Credentials read from env vars only; .env.example provided.
• Radio data is synthetic (OpenF1 team_radio is audio-only); the loop is identical for real transcripts. README carries this + the optimizer-scope caveat.
• Indexed in use-cases/README.md.