feat: add run.sh to f1_radio_rag

[LeoRoccoBreedt]

Summary

Adds CI support to use-cases/f1_radio_rag — the first example to go through the new PR test workflow (from PR #16).

What changed

• run.sh — runs ingest then ask, logging one RAG trace to Opik. Lightweight alternative to the full eval suite for CI.
• config.py — reads OPIK_EXAMPLES_MODEL so CI routes to a cheap model (openai/gpt-4o-mini). Removes LLM_READY guard — missing keys now fail loudly via litellm rather than silently skipping.
• cli.py — removes the LLM_READY dry-run branch in ask.

Tested locally

Ran bash run.sh with real credentials. Trace appeared in Opik workspace under project f1-radio-rag.

Depends on PR #16 merging first for the PR test workflow to fire.

🤖 Generated with Claude Code

[LeoRoccoBreedt]

View sucessful logs: https://www.comet.com/opik/opik-examples/projects/019efb20-5dd1-75e6-9ecb-ce2f47a71681/logs

[fschlz]

Review against current main (PR #16 merged as f48e7d8, so the CI design this branch was cut from has since changed). This needs a rebase + small rework before it can go green.

1. Drop the .github/** and CONTRIBUTING.md changes — they're already on main

This branch predates the #16 merge and re-adds the CI under the old names:

This PR adds	Already on main as
.github/workflows/example-compliance-check.yml	compliance.yml
.github/workflows/pr-examples-test.yml	pr-test.yml
.github/workflows/scheduled-examples.yml	scheduled.yml
.github/ci-examples.json	scheduled.json
CONTRIBUTING.md (run_examples.sh / "CI always has credentials")	reconciled version

Merging as-is would create duplicate workflows running side-by-side and revert the CONTRIBUTING reconciliation. After git rebase origin/main, the diff should be only the three f1 files.

2. Keep these — they're good

• use-cases/f1_radio_rag/run.sh (already the correct name)
• config.py reading OPIK_EXAMPLES_MODEL — fixes the exact gap where f1 hardcoded the full-price model 👍

3. Restore a dry-run path — otherwise the merged pr-test.yml fails this PR

pr-test.yml now has a secrets-free dry-run job: it runs bash run.sh with no API keys and expects exit 0 (the only signal fork PRs get). This PR removes the LLM_READY guard so ask → rag.answer → litellm runs unconditionally → with no key it raises → set -e → job fails.

Fix: in ask, when no LLM key is set, print the retrieved context + a [DRY RUN] would summarise with <model> line and return (the use-case template does exactly this). ingest should be fine creds-free (local Chroma) — worth confirming.

Verify after rework (no credentials)

cd use-cases/f1_radio_rag && bash run.sh                 # must exit 0 in DRY_RUN
uv run ruff check . && uv run ruff format --check .       # the lint job checks both

With creds set, bash run.sh should log a trace (the live-run job).

[fschlz]

see #19 (comment)