[thekevinbot]

Fixes #1478

What

Walk workers spent nearly all their idle time acquiring the shared open mutex, not waiting on the condvar. Instrumented on a 100k-file tree (18,579 dirs, 32 threads): about 200ms of about 230ms of total worker idle time was lock acquisition. Each heap comparison dereferenced Arc<Slot>.key, memory another core had written, so every sift cache-missed while all the workers queued behind it.

tree_walk.rs only:

• Pending carries an inline u128 rank, an order-preserving encoding of the key's leading bytes. Most comparisons stay in the heap's own memory, and full keys break rank ties.
• A worker registers the directories it found and takes back the earliest in one lock acquisition, not two.

listing.rs and scanner.rs are untouched.

Timings

Paired and alternating against fdfind, median per-pair ratio of (dirsql - startup) / fd, with the machine shared with other agents (load about 10). The query is ./**/*.js against fdfind --glob '*.js'.

tree	main	this PR
100k files, 18.6k dirs (node_modules layout)	1.02-1.07x	0.82-0.86x
1M files, 279k dirs (fanout 16)	1.15x	0.99x
800k files, 18.6k dirs (file-heavy)	0.87-0.89x	0.87-0.92x

./**/node_modules/pkg/*.js against fd --full-path on 100k: 0.89x on main, 0.67x with this PR.

E2E Verification

• Ran e2e suites locally for every affected SDK
• Python SDK e2e: pass (32 impacted tests)
• TypeScript SDK e2e: pass
• Rust core e2e (if applicable): pass
• packages/python/e2e-attestations/<branch>.json receipt written if packages/python changed (just e2e-attest-python)
• packages/ts/e2e-attestations/<branch>.json receipt written if packages/ts changed (just e2e-attest-ts)
• Command(s) run: testing-conventions e2e attest 'uv run python -m pytest tests/e2e/glob_speed_*_test.py tests/e2e/query_contract_test.py tests/e2e/fanout_test.py tests/e2e/use_case_join_test.py tests/e2e/stat_speed_test.py tests/e2e/release_speed_test.py -q' (after maturin develop --release and a release CLI build); just e2e-attest-ts; cargo test --workspace --features cli
• Result summary: Python 32 passed; TS 18 passed; Rust all passed.

Changelog: skip-changelog trailer (internal scheduling, same output).