Test github.collect's drop rules, unreadable coverage and run mapping (see #86) #88

Subcategory: Test coverage & depth 9 → 10. Fixed by open PR #86. Close this when #86 merges.

On main, coverage reports 100%, yet three behaviours of collector/jq_collector/github.py collect are never run by any test. Each is one clause of a multi-line expression, and coverage marks the whole expression covered once its first line runs:

• The drop rules: archived (unless JQ_INCLUDE_ARCHIVED), JQ_IGNORE by full name or bare name, and JQ_PUBLIC_ONLY against private repos. GitLab's equivalents are tested in test_gitlab.py; GitHub's are not.
• A coverage artifact whose report can't be read (coverage_percent returns None). It should give no coverage figure (not 0) and keep the artifact id.
• Turning runs into WorkflowRuns: name fallbacks, durations, negative clock skew clamped to 0, and the older failure shown over a newer success.

Every StubAPI in the tests returns latest_runs() == [] and a single public, unarchived repo, so none of these paths ever runs.

Done when: a test exercises each of the three through github.collect, with the 100% floor intact. #86 adds those tests to test_github_http.py.