Feat/repo concierge

[ethancjackson]

Summary

• Adds a repo concierge ADK agent (concierge_agent/) powered by gemini-3.1-flash-lite-preview that answers bootcamp onboarding questions via search_repo_catalog → fetch_repo_artifact.
• Ships a pre-built catalog (context/catalog.yaml + 195 artifacts/*.md) so participants work out of the box after uv sync — no build step required.
• Adds 99_repo_concierge.ipynb (markdown-rendered outputs, RUN_AGENT guard, seed questions) and adk run concierge_agent / adk web terminal path.
• Maintainer workflow: uv run python scripts/build_concierge_context.py when library code or notebooks change.

Participant experience

1. uv sync from repo root (existing setup)
2. Open implementations/getting_started/99_repo_concierge.ipynb, set RUN_AGENT = True, ask questions
3. Or: cd implementations/getting_started && uv run adk run concierge_agent

Test plan

• uv run pytest implementations/tests/getting_started/ -q (8 tests)
• make lint / pre-commit passes
• Smoke: run one seed question in 99_repo_concierge.ipynb with Vector proxy credentials
• Smoke: uv run adk run concierge_agent from implementations/getting_started/

[ethancjackson]

Catalog refresh automation (post-workshop follow-up)

The rebuild script (scripts/build_concierge_context.py) is clean but there's no CI to run it automatically on merge to main. The catalog will silently go stale for new notebooks, READMEs, and implementation files —
only new core .py files are caught by the existing test.

Next step: add a GitHub Action that runs the script and bot-commits the result on push to main. ~20 lines of YAML, paths filter on aieng-forecasting/**, implementations/**, docs/**, README.md, AGENTS.md.