fix(aidd-telemetry): keep a worktree's sessions in every report, even after the worktree is removed

[VincentMenard76]

🎯 What & why

A session run in a linked worktree was missing from every report outside that worktree, and lost for good on git worktree remove: the journal lived at the worktree's own root, git-ignored, and the CLI only read the checkout it ran in. Agent runners give every agent its own worktree, so this was the common case.

🛠️ How it works

• One journal per clone, under <git common dir>/aidd/runs/ (repo.cjs runsDir). Outside every working tree: never committed, no branch dirtied, untouched by git worktree remove, and a bare clone has one too. AIDD_RUNS_DIR still replaces it. Decision and refused alternatives: one-run-journal-per-clone.md.
• The CLI finds the same directory from files, not by spawning git (git-common-dir.ts, paths.ts). A parity test runs the hook's git rev-parse and the CLI's file walk side by side over a plain checkout, a subdirectory, a linked worktree and a bare clone's worktree.
• Earlier journals stay counted: the reader also reads the pre-move aidd_docs/runs of every live checkout of the clone, never writes there, and prefers the new copy when one session has both. forget previews and removes those directories too; check names the real directory instead of a hardcoded aidd_docs/runs.
• No worktree axis in the report: only the cumulative figure matters, and task, backlog, flow and agent already answer what a piece of work cost. worktree_id stays in the journal. fix(aidd-telemetry): a session run in a git worktree is missing from every report outside that worktree, and lost when the worktree is removed #932 was amended to say so.

🧪 How to verify

• node scripts/check-tests-leave-git-alone.js -- node --test 'scripts/__tests__/**/*.test.js' and cd cli && pnpm test.
• telemetry-worktree-journal.e2e.test.ts is fix(aidd-telemetry): a session run in a git worktree is missing from every report outside that worktree, and lost when the worktree is removed #932's QA through the built binary on real worktrees, before and after removal.
• Seven independent review rounds, mutation evidence and accepted limits: review.md.
• Also run by hand on Windows with the real hook and the built CLI: a session in a linked worktree journalled under main/.git/aidd/runs/, attributed to its task from the main checkout and from the worktree, and still after git worktree remove without --force.

⚠️ Heads-up

• Release order. The hook ships in the aidd-telemetry plugin, the reader in @ai-driven-dev/cli, released apart. A newer CLI reads both locations; an older CLI with a newer plugin sees no new session. The plugin README now says to update aidd first, and the decision record lists it. If maintainers prefer a transition period where the hook also writes the old location, that is a follow-up.
• Bundle budget raised 734 → 745 KB (check-bundle-size.mjs, registry entry added): next was already at 733.98 KB, this change measures 737.25 KB.
• The decision record is marked proposed, for a maintainer to confirm or overturn.
• Out of scope, tracked on feat(aidd-telemetry): a task declares its backlog once, and every session on it is found #975: the backlog axis still reads none for a task folder that exists only on another worktree's unmerged branch.
• Commits made with lefthook disabled: its Windows hooks are broken (fix(lefthook): Windows/Git Bash pre-commit hooks fail with node syntax error #948). commitlint and the pre-commit checks were run by hand; validate.yml replays them here.

🔗 Linked issue

Closes #932

✅ I certify

• I DO CERTIFY I READ EACH LINE OF THE PULL REQUEST BECAUSE I AM A SOFTWARE ENGINEER, NOT A AI PUPPY.

🤖 Generated with Claude Code