feat(rc): publish measured checkout diff summaries

[AetherAI3]

Summary

• Publish an RC diff_summary at session start and after a code run settles, based on Git porcelain status and staged/unstaged numstat reads. Settled coding worktree snapshots use the launch project's existing RC outbox and live observer.
• Report clean checkouts as zero; omit line counts when Git cannot establish them for binary, untracked, or failed reads.
• Refuse absolute, external, traversal, and malformed paths before the durable outbox enqueue. Bound path lists and aggregate counts; send no hunks or file contents.
• Cover changed, clean, binary, mixed binary/text, external-path, and failed-count cases with real Git checkouts and focused producer tests.

Verification

• npm run build passed.
• RC producer/outbox/count and new checkout tests passed.
• Previous main-aligned head c0636cd: npm run build and the full local suite passed with 2,902 passed, 43 skipped, 0 failed (2,945 total).
• Current main-aligned head cd85030 includes the RC JSON handoff from main. npm run build and 95 focused RC tests passed. The serialized full suite (AETHER_ATS_PYTHON=<installed Python 3.11> node --test --test-concurrency=1 --test-isolation=none "dist/test/**/*.test.js") recorded 2,902 passed, 43 skipped, 3 failed (2,948 total). Failures were in unchanged controlled Edge and managed preview tests on this loaded Windows host. The preview test passed in isolation; two Edge cases remained flaky in an isolated rerun. This head does not have a clean full-suite result.
• Required GitHub Actions CI and CodeQL jobs did not start: GitHub reports the repository account is locked due to a billing issue. All required checks must run and pass before merge.

Issue: #218