fix(code-index): end every seat wait through one authority

[ScriptedAlchemy]

Refs #2405, #2365, #2550, #2612

Recovered from the audit-leftovers/scheduler-signals lane, which died in the 2026-10-01 restart; re-verified after rebasing on 134157d. The lane's #2209 (MemoryRefusalRetryV1) item was not started and is not in this PR.

Changes

• Store-lock-refused first open waits on the release (fix(code-index-runtime): retry generation reads a store writer blocks #2405 pattern). A cold mount whose first open_with_policy is refused by a contended store lock now waits for the holder's release, a kernel lock wait on a dedicated thread raced against the reservation's cancellation, then reopens. Before, it surfaced as a failed mount. wait_for_code_generation_store_release takes the lock exclusively so a refused writer also outlasts readers.
• One seat-wait authority (fix(code-index): end fresh and ready waits after the graph tail #2365/fix(code-index): verify restarts and wait for the retained seat #2550/fix(daemon): serve cold-daemon reads once their owner is ready #2612 pattern). Every wait for a worktree's seat (decoded generation, current text owner, text serving under a query authority, readiness target) goes through one wait. It subscribes to the root's publication, seat, mount, pass and park signals before the first probe and re-probes only when the registry publishes a change. Its outcome is typed: CodeIndexSeatWaitV1 is seated / parked (CodeIndexSeatParkV1: worker park with remedy, unpublished, publication unreadable, seated-but-not-serving, unreachable readiness) / cancelled / deadline. The advisory runtime's separate wait_for_generation_change loop and the compiler-diagnostics producer's own wait are deleted in favour of it. Seat parks map without a reason table.
• Publication-gate wait wakes on shutdown, not a poll.

Evidence (rebased tree)

• tracedecay-code-index-runtime lib 562 passed (1 ignored), tracedecay-code-index-retention lib 114, tracedecay lib (advisory/deferred/activation selection) 35, full daemon_suite 57 (all pass). New tests: a_first_mount_refused_by_a_held_store_lock_mounts_on_its_release, a_seat_wait_answers_a_parked_worker_with_its_park, a_readiness_wait_ends_when_the_registry_is_cancelled, owner_signals_wake_on_their_own_roots_publication, store_release_wait_outlasts_a_reader.
• I did not build master to run the new tests there. They target behaviour master doesn't have: a store-lock-refused cold mount fails outright on master, and CodeIndexSeatWaitV1 doesn't exist.
• clippy -D warnings (code-index-runtime, code-index-retention, tracedecay, --all-targets, test features), cargo fmt --all -- --check: clean. Windows cross-check: exit 0 (the one BRANCH_TRACKING_FAILED dead-code warning is pre-existing).
• Known limit: on shutdown the release-wait thread is detached, not interrupted. It ends when the holder releases or the process exits.

[changeset-bot]

⚠️ No Changeset found

Latest commit: c5137f2

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR