fix(loro.js): apply Text/List deletes imported while detached

[zxch3n]

Summary

A Text/List delete imported while a loro.js document was detached was lost when the document moved to a version that includes it:

doc.checkout([{ peer: "2", counter: 1 }]);            // "he"
doc.import(sourceUpdateThatDeletesTheO);
doc.checkoutToLatest();   // main: "hello"; loro-crdt and a fresh import: "hell"

The same happens for checkout(frontiers) to any version that contains the delete, and for lists. Afterwards the replica disagrees with every other peer. This was one of the "peers do not converge with a replay of the same updates" classes found by fuzzing.

Rationale

Importing while detached only integrates the records into the history; the container state (including the sequence's deletion index) is updated by the next transition. #canTransitionRecords already refuses the incremental transition when an insert is not materialized in the sequence, which falls back to a replay. It accepted any delete, though. #applyVersionTransition then looked the delete up with idRunsDeletedBy, found nothing, and skipped it. The fix makes #canTransitionRecords require that the sequence has recorded every element a delete removes (idRunsDeletedBy covers the op's full length). Otherwise it falls back like the insert case. Deletes integrated normally keep the incremental path; the extra check is O(log n + runs) per delete op in a transition, the same lookup the transition itself does.

Map, Tree, and MovableList transitions read per-subject histories that are indexed on integration even while detached, so they were not affected.

Validation

• New tests in loro-js/tests/runtime.test.ts: text and list, via checkoutToLatest() and via checkout(oplogFrontiers). All four fail on main, and the expected values equal loro-crdt (WASM build of this main) and a fresh import.
• Differential fuzz (harness in test(loro.js): add a multi-peer differential fuzz harness #1141): with this change, the remaining peer-vs-replay and checkout-vs-replay failures in 500 Map/List/Text/Counter seeds (after fix(loro.js): stop a concurrent insert at the end of its origin's subtree #1139) disappear.
• pnpm --dir loro-js format:check, lint, typecheck, test (251 passed), build.
• bench:complexity (1k–8k), 7 alternating rounds vs main, loadavg 12–13: 284 items, none both >1.5× and >0.3 ms slower. Total 922 ms on main, 938 ms here. (An earlier run at loadavg 40 with a concurrent fuzz job flagged unrelated map-import items that do not reach this code; they did not reproduce.)

🤖 Generated with Claude Code

[github-actions]

WASM Size Report

• Original size: 3210.11 KB
• Gzipped size: 1062.59 KB
• Brotli size: 744.13 KB

[zxch3n]

Closing: covered by #1126 (5e409d1d), which added the same check to #canTransitionRecords (a delete op must have recorded every element it removes, otherwise the transition replays).

Checked on main at c0ff1255: text and list deletes imported while detached are applied by both checkoutToLatest() and checkout(oplogFrontiers) (hell / ["h","e","l","l"], same as loro-crdt). The 4 regression tests from this PR pass on main.