fix: move Darrow state into user directories

[BjRo]

Why

Closes #225. Review artifacts stored under repository Git directories have an unclear lifetime. Put Darrow state under one user-local root and make review cleanup explicit. This also updates the default UV project-environment location from the already implemented #226.

What changed

• Store review runs under ~/.darrow/reviews/ on macOS/Linux or %LOCALAPPDATA%\Darrow\Reviews\ on Windows, with repository/worktree buckets, unique runs, and private permissions where supported. Keep DARROW_REVIEW_STATE_DIR as an absolute-path override.
• Add 30-day dependency-aware pruning and pin/unpin commands. Preserve review evidence referenced by newer runs or pinned by a user; retain blocked review results so they can be pinned.
• Move the default UV project environment for all 12 registered Python plugins to ~/.darrow/cache/ or %LOCALAPPDATA%\Darrow\Cache\. Preserve DARROW_CACHE_DIR; leave UV_CACHE_DIR under UV or caller control.
• Update the review invariant, Python quality spec, plugin documentation, and colocated evals for the new paths and cleanup rules.

Verification

• bun run check:python passed across all 12 packages; darrow-review: 229 tests, 96.82% line and 95.73% branch coverage.
• bun run lint, bun run check:docs, bun run lint:ts, bun run typecheck, and git diff --check passed.
• Eval runner and fixture tests: 36 passed; bun run test:eval-runner-compatibility: 5 passed.
• Review backend frozen fresh-install test passed for all 7 entrypoints. Skill inspection passed. Focused Codex code-review-empty-diff live eval passed task and activation (1/1); no matched control was required for this focused run.
• Claude live evals were attempted: implicit empty-diff failed activation, and explicit reviewer-route-unavailable activated the skill but did not produce its required artifact commands. These are open skill-behavior verification limits.
• The changed launcher scripts are Python, so bash and /bin/bash script checks do not apply.

Review notes

No migration runs. Files in previous Git and UV locations remain untouched. Review evidence is retained until eligible for 30-day pruning; cache files are disposable and managed separately. Windows behavior has unit coverage with simulated platform paths and locking; it was not run on a native Windows host. Review darrow_review/storage.py and the eval runner state isolation for the highest-risk changes.

Checklist

• I have read and followed CONTRIBUTING.md, including the contribution licensing terms.
• I added or updated the applicable invariant before implementation, or this change does not affect a capability invariant.
• I added or updated colocated evals, or this change does not affect skill behavior.
• I confirmed that each changed plugin remains self-contained, or this change does not affect plugin content.
• I ran bun run check:python, or this change does not affect registered Python packages or their repository quality infrastructure.