Count only the work that matters: recency, mechanical commits, mirrored test pairs, logical changes #57

Hotspots, coupling, cohesion and trends should describe the work a team does now, not a repository's setup phase or mechanical commits.

• Recency: weight revisions by age (half-life, default 6 months, --half-life, 0 disables). Keep raw revisions; scores use weighted revisions.
• Mechanical commits: pure renames (R100) and mode-only changes, whitespace-only diffs, revert pairs inside the window, patch-identical duplicates, and commits listed in .git-blame-ignore-revs no longer count as revisions, churn or coupling evidence. The report lists how many of each kind it set aside.
• Mirrored test pairs: src/a/b.ts ↔ test/a/b.test.ts (and tests/, __tests__/, spec/, e2e/) are test pairs.
• Logical changes: commits of one pull request (squash suffix (#123) or one merge) or one ticket key count as one change for coupling; fall back to single commits and say so.
• Bugs: hub detection treats test step files and mocks under test directories as hubs; a tracked manifest that matches a .gitignore pattern is ignored by package detection; a workspace package that owns no files itself is not resolved by name.

Done when each behavior has a test that fails without it, the README explains the new counting rules, and the effect is measured on several real repositories.