perf(shared-fs): trim redundant batch result assembly

[peerbit-org]

Summary

Remove result-only work from each applied writeBatch edit. The new version includes every captured head as a parent, so the old-head filter has no surviving entries. Pass the existing FileVersion directly to the formatter, as writeFile already does.

This removes two temporary row objects, two parent-list JSON decodes, a Set and a redundant old-head scan per applied edit. No dependency, trust, live namespace lookup, publication order, manifest, persisted receipt or post-version chunk-presence repair changes. No end-to-end speedup claim.

Validation

• Six focused cases cover exact stored/returned fields for create, overwrite and multi-head conflict merge; unchanged/delete result slots; and independent result-only equivalence/work counts for 0/1/64 parents.
• Full first-attempt strict library suite: 525 passed, 10 skipped, 0 retried tests, 0 missing instrumentation, 64.00s, natural process exit 0. Original assertions and embedded retry declarations were not changed; the strict reporter would fail any retry. Raw SHA256: 9472e470b864a8cbd7734320e5c64d65f4d14f71fb808a4e61b1139ccc91477f.
• Library build, source/focused-test typechecks, lint and formatting pass.
• npm dry-run package: 83 files, 2,205,055 unpacked bytes, 493,449 compressed bytes; no tests shipped, existing size budget respected.

Development fixture mistakes (history enumeration order and test-only typing) were corrected before final validation; no product timeout or retry policy was changed. The first full strict run passed; it was not rerun.

This is based on master b2c9d300, separate from held trust PR #344 and held cleanup/upgrade branches. A separate 16-worker comparison of baseline versus #344 found batching faster in 12/16 matched cases, including counterexamples on baseline; that experiment does not measure this cleanup or clear #344's old strict failures. Keeping this draft pending fresh CI/raw retry audit; no auto-merge or release.

[peerbit-org]

First-attempt raw-log audit for exact head 8983eba50531d1abc1bed89baeebb44e7981b68b, run 33975452826 attempt 1: ordinary CI is green, but this is not a clean zero-retry signoff.

• Ubuntu: durable-disposal.test.ts naming-conflict/tombstone disposal case used (retry x1) (110555ms).
• Windows: persistent-multi-writer.test.ts three-authenticated-replica lifecycle used (retry x1) (184419ms).
• macOS had no retry marker. All three eventually passed 525 library tests (10 skipped), 33 CLI tests and the six new result-assembly cases. Interop and package checks passed.

Neither original first-attempt failure stack is exposed by the ordinary reporter. Nearby snapshot-publication stderr is context, not established cause. These are old-cohort jobs, not evidence against the upcoming upstream release. No job was rerun and no timeout/assertion changed.

Keeping this draft held. The isolated next-cohort integration branch includes the existing strict first-failure reporter, the durable-readiness upgrade, trust-cache fence and this small cleanup. It will receive only the verified published cohort and fresh strict cross-platform validation; it does not overwrite the original checkout or held PR worktrees. No auto-merge or release.