feat(review): the pin follows the lock

[haribo]

Summary

• ADR 0024 (let the pin follow the lock instead of the review status #248, validated): the displayed edition is derived, never stored — a live lock shows the edition stamped at claim, a case nobody holds always reads at the latest; cases.current_edition_id is gone with AdvanceCurrentEdition and ReleaseToLatestEdition, and intake stops skipping to-review cases (the dd9c192d95e0 freeze: evidence hidden to protect a reviewer who did not exist)
• One resolver behind every read and stamp: displayedEdition (lock → latest) feeds factsOf, the grid's resolveEdition, comment anchors, reference stamps, coverage restores
• A delivery still advances the case at once (a delivery advances the case onto the edition that carries it #142): deliver re-stamps the live lock onto the latest — nothing to do when nobody holds
• The audit's gap closed: when a review settles, the state re-derives against the latest edition, past the saver's own lock — otherwise a case could stamp accepted blind to an edition already waiting
• One rule added during implementation, flagged for review: a claim is fresh (on-load — re-stamps onto the latest) or a heartbeat (keeps the bytes). Reloading is leaving and coming back; without the distinction a reload raced its own keepalive release and showed one edition or the other by timing. In the ADR, in the contract (POST lock {fresh}), in the client
• Migration 00022 (case_locks.edition_id, drop of the stored pin), exercised Up→Down→Up

Test plan

• The issue's two validation tests: TestAFreeCaseReadsAtTheLatestEdition (the dd9c192d95e0 shape — a video landed after birth shows), TestAHeldCaseKeepsItsBytesUntilTheLockDies (held bytes stable through a push, latest after expiry, no write in between)
• TestADeliveryAdvancesTheCaseOntoItsEdition green through the re-stamp, untouched
• grep -rn current_edition_id apps/server/ answers only the migrations
• Full gates: gen-check, be-check, be-test (12), db-test, fe-check, fe-test (110), fe-test-e2e (64)

Existing tests reworked (the issue pre-approved the list; two more surfaced):

• TestAnIntakeDoesNotMoveTheBytesUnderAReviewer — now holds the lock, as the guarantee requires
• TestACaseCatchesUpOnceItsReviewEnds → TestACaseCatchesUpAsSoonAsNobodyHoldsIt
• TestAnOlderEditionDoesNotDrawAStepBornLater needed nothing: it pins through the explicit edition parameter, which survives
• Surfaced by the run: TestAnOlderEditionCanStillBeRead's second half carried the status premise (default read frozen) — it now asserts the free case reads the latest; TestACommentAnchorsToTheCaptureItWasWrittenAbout's fixture passes the anchor edition AttachCommentVariant now requires

Closes #248