STATUS-011: status re-hashes every clean tracked file — add the wtlog mtime-stamp fast path

Now: OPEN

Status's classifyMerge calls wtEqBase on EVERY in-base+on-disk path (be/shared/classify.js:417), a full open/readAll/blob-hash per file, even when the file is untouched since checkout. Yet every verb-written mtime is already recorded in the wtlog stamp-set (wtlog.js has(ts); get/post stamps, DIS-057 patch bands), and stage.js:150 already uses exactly this fast path: mtime ∈ stamp-set → clean, no read. Status is the outlier: on a big wt the content sweep dominates wall-clock (PACK-003 fixture: full re-read+re-hash of every path, no stat shortcut at all).

Input

Context

jab status wall-clock on a large wt is dominated by wtEqBase content reads; the wtlog already carries the ground truth for "untouched since the verb wrote it".

classify.js:417 — wtEqBase(wtRoot, path, b.sha) unconditionally; classify.js:127-161 is the per-file open/readAll/frameSha cost.
wtScan already lstat's every path and carries ts (mtime ron60 BigInt, JS-042) in the merge row — the stamp is in hand, unused.
stage.js:150,216,242 — put/post staging already skips content compare when mtime ∈ get/post stamp-set; status never got the same treatment.
DIS-023 ruled clean must be content-confirmed, never mtime-ALONE — that ban targets the old re-stamping drift gate, NOT a read-only membership test against verb-recorded stamps (which stage.js has used since).

Goals

Fast path in classifyMerge: for an in-base+on-disk path, if wt mtime ∈ the wtlog stamp-set → treat as eqBase true (bucket ok) without opening the file; only mtime-miss paths pay the read+hash.

Same stamp sources stage.js trusts: get/post row stamps + DIS-057 patch bands (a patch-band stamp still routes pat/mrg/cnf first, as today).
Status stays read-only: membership test only, NO re-stamping (the DIS-023 sin).
Listing mode (eq rows, JAB-018) keeps emitting the wt mtime it already has.

Constraints

Byte-parity with today's output on stamp-miss paths; the fast path may only skip work, never change a bucket vs content truth for verb-stamped files.
A restored/copied mtime that COLLIDES with a recorded stamp over edited bytes reads clean — accepted risk, same trust stage.js already extends (ron60 stamps are sub-ms; collision is adversarial, not accidental).
No new index/cache files; the wtlog stamp-set is the only source.

WIP

Design decisions

3-line fast path in the inBase+onDisk arm: !pb && w.ts && wtlogReader.has(w.ts) short-circuits wtEqBase; no reader threading needed (wtlogReader already in scope).
The !pb guard is load-bearing: the cnf band slot IS the patch row ts, a stamp-set member — pb routes con/mrg first.
Membership test only, zero writes — DIS-023-safe; the stamp-collision misread stays the stage.js:150 accepted risk.
Repro test/statusfast.js hooks io.open to count content reads: clean+stamped wt reads 0 (was N), stamp-miss paths keep the content sweep + byte-identical buckets.

TODOs

Thread the wtlog reader (or its has(ts)) into classifyMerge's inBase+onDisk arm; consult before wtEqBase at classify.js:417.
Repro/perf test first: N-file wt, all clean via get; assert status does zero content reads (count io.open via hook or timing bound) and output is byte-identical.
Verify the patch-band interplay: pat/mrg/cnf stamps must not collapse to ok (pStamp check precedes the fast path).
Check the con-row marker scan (classify.js:383) — already gated on conSet, unaffected.

Blockers and bummers

get never restamps checkout mtimes (no io.setMtime in get.js/checkout.js) — only post/put/patch stamp, so a FRESH CLONE still pays the full content sweep; stamping get checkouts would need its own ticket.