fix: harden diff calc for shallow text histories

[lodyai]

Summary

• Add an LCA-aware diff_calc fuzz target plus focused regression coverage for shallow text, state-only, tracker, and LCA cases.
• Fix shallow text tracker seeding and partial-delete handling when only part of a CRDT id span is active.
• Clamp shallow snapshot/LCA traversal around trimmed history, and include target state delta in state-only shallow snapshot exports to avoid replaying text ops in an invalid context.

Performance

Quick Criterion comparison against origin/main after rebasing:

• import_regression/text_split_checkout_1024: origin/main 4.3524 ms, branch 4.5337 ms, about +4.2% median.
• import_regression/list_checkout_4096x1024: origin/main 4.7133 ms, branch 4.9708 ms, about +5.5% median.

This was a short run with 10 samples, 1s warmup, and 3s measurement time, so treat it as directional. The list case intervals overlap; this does not look like a large regression, but the text checkout path is a little slower in this sample.

Validation

• cargo fmt --all --check
• cargo test -p fuzz diff_calc --test test -- --nocapture
• cargo fuzz run diff_calc fuzz/artifacts/diff_calc/*
• cargo fuzz run diff_calc ~/loro-fuzz/corpus/diff_calc -- -max_total_time=18000: 304119 runs in 18001s, no crash. This long run was before the final conflict-free rebase; post-rebase focused tests were rerun.
• cargo test -p loro state_only -- --nocapture
• cargo test -p loro-internal common_ancestor --lib -- --nocapture
• cargo test -p fuzz one_doc_tree_move_before_cycle_error_is_ignored --test test -- --nocapture
• cargo check -p loro-internal

[github-actions]

WASM Size Report

• Original size: 3125.98 KB
• Gzipped size: 1004.43 KB
• Brotli size: 699.01 KB