refactor: split scan_repo, _folder and _entry below CC 15 (#51)

[tschm]

Fixes #51.

Acceptance criterion (verbatim): "radon reports no block above CC 14 in jq_collector, and coverage stays at 100%."

What changed

This is a pure extraction: no behaviour change, and every error and log message is byte-identical.

• localgit.scan_repo, CC 18 → below C. The status counts, HEAD time and sha, branch naming (including detached HEAD) and stash count each moved into their own helper: _status_counts, _head, _branch_name, _stash_count. One side effect: git stash list now runs a few calls later within a single scan. It's read-only and independent of the other calls.
• repos._folder, CC 17 → below C. Pulled out _refuse_single_repo_keys, _checkouts_in (the unreadable and empty folder refusals), _warn_spent_excludes and _log_taken.
• repos._entry, CC 17 → 13. Pulled out _remote_only (the entry has no path:) and _no_checkout (the path: has no checkout).

uvx radon cc collector/jq_collector -s -nc afterwards: the highest is 14 (repos.load, localgit.scan, github.collect, GitHub.active_workflows), and nothing is above 14.

Gates (run locally on this branch)

• ruff check and ruff format --check: pass
• mypy (strict): no issues in 11 files
• pytest --cov=jq_collector: 456 passed, 100% of lines and branches (402 branches, the same count as before)
• pytest on Python 3.11: 456 passed
• prek run --all-files: everything passes except bandit and shellcheck. Those two are already red on main and are fixed by fix: run the pre-commit hooks in CI and clear bandit and shellcheck (#59) #63, not by this PR.

Not addressed

• This touches localgit.py, as fix: run the pre-commit hooks in CI and clear bandit and shellcheck (#59) #63 does. They edit different hunks (_git and the import vs. scan_repo), so they should merge cleanly in either order.

Closes #51

🤖 Generated with Claude Code