…hase 5)

Phase 5 (final) of the autolens_profiling z_feature roadmap. Wires up
two GitHub Actions workflows + threads AUTOLENS_PROFILING_SMOKE=1 into
every profile script so the lint workflow's smoke step short-circuits
cheaply.

What lands in this PR
---------------------

1. ruff.toml — repo-root ruff config (the repo isn't a Python package
   so there's no pyproject.toml). Conservative rule selection:
   E/F/W/I/UP/B with E501/E402/F401/B008 ignored for scientific code.

2. .github/workflows/lint.yml — Workflow 1, PR + push-to-main gate.
   Target wall time <5 min on CPU-only ubuntu-latest. Steps:
   - ruff check . + ruff format --check .
   - python scripts/build_readme.py --check (dashboard idempotence)
   - lychee markdown link-rot across every README.md
   - Smoke: one script per section with AUTOLENS_PROFILING_SMOKE=1
     (catches import-graph breakage without running the full profile)

3. .github/workflows/profile.yml — Workflow 2, manual + on release.
   - workflow_dispatch with optional `sections` input
   - release: published trigger
   - Runs every script under likelihood/, simulators/, searches/nautilus/
     with continue-on-error per section (single regression doesn't block
     the dashboard refresh for the other 16)
   - Skips simulators/point_source.py in the loop (its default
     dataset_name="simple" overwrites Phase 1's tracked likelihood JSONs)
   - Runs scripts/build_readme.py to refresh dashboard tables
   - Commits diff back to main as github-actions[bot] with [skip ci]

4. .github/workflows/README.md — documents both workflows + design
   decisions captured for future maintainers.

5. AUTOLENS_PROFILING_SMOKE=1 short-circuit threaded into 17 scripts:
   - likelihood/imaging/{mge,pixelization,delaunay}.py
   - likelihood/interferometer/{mge,pixelization,delaunay}.py
   - likelihood/point_source/{image_plane,source_plane}.py
   - likelihood/datacube/delaunay.py
   - simulators/{imaging,interferometer,point_source,cluster,group,multi}.py
   - searches/nautilus/{simple,jax}.py
   Inserted via AST-driven helper at the first non-import top-level
   statement of each script (after the module docstring + initial
   imports, before any module-level execution). Verified end-to-end:
   AUTOLENS_PROFILING_SMOKE=1 python <script> exits in <1s for all 3
   representative scripts smoked locally.

Design decisions resolved
-------------------------

- Runners: CPU-only on github-hosted ubuntu-latest. Self-hosted GPU
  runners can be added later as a separate job without restructuring.
- Cadence: workflow_dispatch + release-triggered only. No weekly cron.
- Bot identity: github-actions[bot] with [skip ci] subject line.
- Failure handling: continue-on-error per script with ::warning::
  annotation; dashboard cell shows ERR rather than blocking refresh.
- Smoke env var name: AUTOLENS_PROFILING_SMOKE=1 (proposed in the
  Phase 5 prompt, kept as-is).
- ruff config location: ruff.toml at repo root rather than
  pyproject.toml — this isn't a Python package, ruff supports the
  standalone config file natively.

Test plan
---------

- [x] yaml.safe_load PASSES on both lint.yml and profile.yml.
- [x] py_compile PASSES on all 17 SMOKE-instrumented scripts.
- [x] AUTOLENS_PROFILING_SMOKE=1 python likelihood/imaging/mge.py exits
      0 in <1s with "[smoke] ... imports + module setup OK; exiting."
- [x] Same for simulators/imaging.py and searches/nautilus/simple.py.
- [x] scripts/build_readme.py --check exits 0 after SMOKE insertions
      (dashboard idempotence preserved).
- [ ] First lint workflow run will be the real check — local YAML
      parses but actual GitHub Actions semantics validate on the
      runner. Any failures from there get follow-up PRs against this
      workflow file.

Closes #7.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>