fix(loro.js): root shallow snapshots at the latest critical version

[zxch3n]

Summary

loro.js rooted a shallow snapshot at the requested version, or at the meet of its heads. When a branch merged after that version had forked below it, a retained op was concurrent with the root. loro-crdt roots the snapshot at the latest single-head critical version of the requested frontiers and the latest version (#1095, calc_shallow_doc_start).

Effects on main:

• The root differs from a Rust export of the same document (e.g. 2@1 instead of 0@1 for the "past version with a later branch below it" case in crates/loro/tests/shallow_root_critical.rs).
• Checking out retained versions of such a snapshot and returning to the latest version could reorder concurrent Text/List elements. This is the "shallow snapshot of concurrent history differs from the full-history document" failure from random fuzzing. On a branch with the other Text/List fixes from this session, Map/List/Text/Counter fuzzing with the snapshot and shallow oracles went from 175 failing seeds out of 200 to 3 out of 300 with this change. The remaining 3 are the multi-op root commit case that fix(loro.js): keep root-time map values and tree nodes when a shallow doc retreats #1127 fixes (verified on the fix(loro.js): drop containers deleted before the shallow root, keep tree node metadata #1128 head plus this change).

Rationale

Port of Rust's latest_single_head_critical_version (dag.rs) as #latestSingleHeadCriticalVersion. It is a max-heap descent over change spans from both frontier sets, with the same coalescing, lamport-based trimming, implicit same-peer predecessor, and empty-version fallback when a chain reaches a root. #calculateShallowStart now calls it with the latest history frontiers. The StyleStart nudge and the clamp to an existing shallow root are unchanged. The walk only runs on shallow export and is bounded by the history between the requested version and its critical version, as in Rust.

Validation

• New loro-js/tests/shallow-root-critical.test.ts: ports of independent_heads_have_no_critical_root, shared_root_multi_head_still_trims, and past_version_with_later_branch_below_it; a concurrent-list case; and a randomized check (40 histories × several targets) that the root is critical and that the shallow doc matches the full doc at every retained version and after returning to latest. Three of the five tests fail on main. The expected roots are the ones loro-crdt (WASM build of this main) picks for the same documents.
• The fuzz harness (test(loro.js): add a multi-peer differential fuzz harness #1141) compares both implementations' shallow roots after every scenario; with this change they agreed in every comparison of the 300-seed run.
• pnpm --dir loro-js format:check, lint, typecheck, test (252 passed), build.
• bench:complexity (1k–8k), 7 alternating rounds vs main, loadavg 24–31: 284 items, none both >1.5× and >0.3 ms slower. Total 1219 ms on main, 1172 ms here.

🤖 Generated with Claude Code

[github-actions]

WASM Size Report

• Original size: 3277.59 KB
• Gzipped size: 1089.33 KB
• Brotli size: 762.83 KB

[zxch3n]

Merged the latest main (c0ff1255, includes #1126–#1128, #1136, #1137); no conflicts. pnpm -C loro-js check with LORO_REQUIRE_WASM_REFERENCE=1 passes (26 files, 419 tests).

Still needed on current main: with the differential harness from #1141 (Map/List/Text/Counter, snapshot + shallow + checkout oracles, 200 seeds, shallow-root comparison disabled so data errors are visible), main fails 124 seeds. Seven of them are shallow-latest: the imported shallow doc's latest state is wrong. The rest are wrong values after checking out retained versions. This branch fails 0 of the same 200 seeds.