[thekevinbot]

Refs #1461

Step one of the wide-directory redesign: read a directory's listing in chunks. This is a structural change with no observable effect. Later steps sort and judge each chunk while the next one is being read, then merge the chunks.

• listing::read_chunks(dir, emit) hands a directory's entries to emit as Chunks, at most CHUNK_ENTRIES (8192) at a time, as it reads them. It applies on every platform (Linux RawDir, other Unix Dir, Windows read_dir).
• Each chunk has its own name buffer. A Listed entry records its chunk and u32 offsets into it, so it is self-contained across chunks and shrinks from 32 to 24 bytes.
• Listing::read collects the chunks and sorts them together, so nothing outside listing.rs changes.
• The branch carries the test: fix main's lib-test compile (scan_glob_rows visit argument) #1527 fix commit (identical change, merges cleanly) so its CI could compile the lib tests while main was red.

Output is byte-identical: same md5 for ./docs and ./ on a 1.6M-item tree.

Timings

These are paired medians on ZFS with other agents running, load 10-34.

case	before (main)	after
./docs (1.57M entries) vs rg	0.74-0.85	0.79-0.87
./* vs rg	0.70	0.80
after/before direct, ./docs		0.989
after/before direct, ./*		1.009
after/before direct, 20k dirs ./**/*.md		0.967

Neutral within noise. Where a wide directory's time goes (instrumented, ./docs): getdents takes about 350ms, pushing entries about 25ms, sorting about 55ms, and judging about 200ms. Sort and judge are what the later steps hide behind the read.

E2E Verification

• Ran e2e suites locally for every affected SDK
• Python SDK e2e: pass (targeted)
• 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: uv run maturin develop --release, cargo build --release -p dirsql --features cli, testing-conventions e2e attest 'uv run python -m pytest tests/e2e/glob_speed_one_level_test.py tests/e2e/glob_speed_recursive_test.py tests/e2e/glob_speed_depth_bound_test.py tests/e2e/glob_speed_gitignore_test.py tests/e2e/glob_speed_dotfiles_test.py tests/e2e/glob_speed_matchers_test.py tests/e2e/glob_speed_node_modules_test.py tests/e2e/glob_speed_symlink_test.py tests/e2e/stat_speed_test.py tests/e2e/query_contract_test.py -q', just e2e-attest-ts, cargo test --workspace --features cli
• Result summary: python 26 passed. A first run had one failure in glob_speed_symlink ./docs/, a 12ms native case, at load 15; it passed on rerun and in the full rerun. ts passed; cargo workspace all passed.

Changelog / Migrations

• skip-changelog trailer: internal restructuring, no observable change.

Merge with main (2026-10-19)

• Merged main after perf: walk drops unmatchable file names early and reads ahead in walk order #1521 and perf: judge a wide directory in parts the walk schedules like directories #1535: Linux reads through Dir again, so read_into (now Dir on every Unix) takes a Chunker; RawDir and ENTRY_BUFFER are gone. Listing::holds stays.
• judge now skips an entry with an empty name. An in-diff mutant that empties every name made the walk re-enter each directory forever and killed the mutation runner.
• Re-measured on the merged head against main, interleaved medians: ./**/*.md 1.018x, ./**/{a,b}*.md 1.006x (neutral). Python e2e subset passed, 29 tests.