feat: doctest the forge and origin helpers, and run doctests in CI (#40)

[tschm]

Fixes #40

Acceptance criterion (verbatim):

each of forge.py and origin.py carries at least one doctest, and CI executes and passes them.

What changed

Examples. They go where the docstrings already make claims, so each one checks something the prose asserts:

• forge.normalise_gitlab_status: failed → failure, case and whitespace ignored, manual → cancelled, and an unknown status → stale (the "never a fleet-wide red" promise).
• origin.forge_for_host: gitlab.com and gitlab.* → gitlab; anything else, including an empty host → github.
• origin.parse: the scp form; a nested GitLab namespace keeping every segment (the bug the module docstring describes); a path remote keeping only its tail; a one-segment path → None.

Wiring. collector/pyproject.toml adds jq_collector to testpaths and --doctest-modules to addopts. ci.yml is unchanged: its existing test step in collector (lint + tests), and the collector (tests, Python 3.11) job, now collect the doctests automatically. The same goes for a local uv run pytest.

Checks (run locally from collector/, as CI does)

• uv run --frozen ruff check … / ruff format --check …: pass
• tests on 3.12: 450 passed (447 + 3 doctest items: forge.py: 1, origin.py: 2), 100.00% coverage, floor met
• tests on 3.11: 450 passed, 100.00%
• Negative check: in a scratch copy with one expected output changed ('failure' → 'failed'), pytest exits 1 with FAILED jq_collector/forge.py::jq_collector.forge.normalise_gitlab_status.

Not addressed

Doctests for the other nine modules. The issue scoped this to the two pure helper modules; the rest mostly need a network client or a git checkout to say anything.

Merge note

#44 (the mypy gate) also edits collector/pyproject.toml, but in different hunks (the dev group line, and a new [tool.mypy] table). The two should merge in either order without conflict.

Closes #40

🤖 Generated with Claude Code