docs: document how to run the collector's CI checks locally (#47)

[tschm]

Closes #47

Acceptance criterion (verbatim): "someone new can reproduce every CI check locally from the README without reading ci.yml."

What changed: a "Development" section in collector/README.md that lists every check CI runs:

• collector/: lint, format, types, tests (with the 100% floor explained, since it comes from config rather than a flag) and the Python 3.11 floor run.
• Repository root: the dashboard checker, the shell-script parse, the compose parse, the image build and the book's mkdocs build --strict.
• The image smoke test: described in words rather than transcribed, because it's a 40-line script. make up is given as the local equivalent, and the section points to the exact assertions in ci.yml.

This goes beyond the list in the issue body, which left out the image job and book.yml. The criterion says every check, so they're in.

It also notes that the 3.11 run rebuilds .venv, so the two test runs go one after the other.

Gates: I ran every documented root-level command from this branch: bash -n on image/*.sh, docker compose config --quiet and mkdocs build --strict all pass. The collector commands are the ones CI runs unchanged. The README passes markdownlint (MD013 disabled, as in the org's pre-commit config) and its 2 bash fences parse. docker build wasn't re-run, since the command is copied verbatim from CI and no image input changed.

🤖 Generated with Claude Code