Reconcile Tailtest install and hook docs

[BananaAccurate]

Root cause

Tailtest documentation still described the old marketplace/direct-clone boundary and contained stale historical validation-count claims after the runtime/hook contract stack was repaired.

Change

• Makes marketplace install the recommended current path:
  • codex plugin marketplace add avansaber/tailtest-codex
  • codex plugin add tailtest@avansaber-tailtest
  • start a new Codex session and review/approve Tailtest SessionStart, PostToolUse, and Stop hooks when prompted.
• Keeps direct clone as the fallback path through scripts/init.sh.
• Documents that marketplace/plugin-bundled hooks do not require per-project .codex/hooks.json on current Codex.
• Reconciles README, AGENTS, RUNTIME, CONTRIBUTING, and CHANGELOG with the current SessionStart, PostToolUse, Stop, safe-defer, and untrusted-data contracts.
• Removes stale unsupported historical test-count claims; README now labels current branch validation evidence and says macOS was not rerun for this docs update.

Safety boundaries

• This branch is stacked on PR Align PostToolUse payload/output contract #14 / Integrate safe Stop defer boundaries #13 / Harden Tailtest hook runtime contract #12 / Add CI and satisfy Tailtest validation gates #11 / Fix stale Tailtest mtime queueing #10 / Fix compact watermark rebasing and Windows test stability #8 because upstream main lacks the prerequisite Tailtest repairs and CI workflow.
• It does not include PR Honor explicit no-more-tools Stop boundaries #9.
• It does not touch runtime Python or hook JSON.
• It does not claim an upstream merge, released artifact, green upstream checks, or release completion.
• Commits use the GitHub noreply identity.

Validation

Validated at head c0d11e860c604f3027a3f02ea362e11c97fef68d:

• Documentation count scan: only current README branch evidence remains.
• Windows: python -m pytest -q — first run exposed a non-reproduced TestGitCleanMtimeChurnSkipped::test_clean_tracked_file_with_new_mtime_is_skipped_when_git_required failure.
• Windows focused reruns:
  • exact failing test — passed.
  • tests/test_mtime_sweep.py — 50 passed.
  • full python -m pytest -q rerun — 437 passed.
• Windows: python -m ruff check . — passed.
• Windows: python -m ruff format --check . — passed.
• Windows: git diff --check — passed.
• WSL/Linux: python3 -m pytest -q && python3 -m ruff check . && python3 -m ruff format --check . — 436 passed, 1 skipped; Ruff passed.

Task review approved after three scoped fix rounds.

Remaining limitations

• The initial non-reproduced Windows mtime/Git-clean failure remains evidence for the later runtime-regression/audit slice.
• Upstream repository currently reports no PR checks because workflows are absent from upstream main; PR Add CI and satisfy Tailtest validation gates #11 adds CI.
• Merge, tag, and release remain blocked until an upstream maintainer or an authorized integration performs those actions.