perf: stop splitting decompressor input into 16 KiB writes

[gmaclennan]

Entry data was split into 16 KiB writes before decompression to cap WebKit's output chunks, since WebKit inflates a whole input write into one chunk. That capped a single chunk at about 16 MiB instead of 64 MiB, which only matters for data near deflate's 1032:1 limit such as zip bombs or long runs of zeros, and the size check already rejects an oversized chunk before it reaches the caller. Rather than work around a WebKit limitation, this removes the split. Reads from the source are already at most 64 KiB, the queue's high-water mark, so that remains the bound on each write.

The split had a real cost. Decompressing 10 MB of incompressible data with 16 KiB writes was 1.7–2.2x slower in Chromium, Firefox and WebKit and about 55% slower in Node, and text was up to 23% slower. On the realistic large and medium benchmarks in #29 it is about 5–10% end to end. The earlier benchmarks in #26 didn't show this because their fixture compressed about 300:1, so entries were too small to be split.

The test that bounded the size of a decompressed chunk mostly checked engine behaviour: Node, Chromium and Firefox emit small chunks regardless of what zip-reader does. It is replaced by a test of what zip-reader controls, using a source that records reads: entry data is read in slices of at most 64 KiB, and only a few slices are read ahead of the consumer rather than the whole entry.

The new test failed on Node 22, and on main's code too, which exposed an existing problem: Node before 24.19 counts each chunk written to DecompressionStream as size 1 against a 16 KiB high-water mark (nodejs/node#62986), so pipeThrough read a deflated entry's entire compressed data into memory ahead of the reader. Entries are now fed to the decompressor one awaited write at a time, which bounds read-ahead to a few chunks on every runtime with no measurable cost. A comment links the Node fix, so the workaround can go when Node 22 support is dropped.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 65.45%. Comparing base (d62e0a6) to head (72bf0e5).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #30      +/-   ##
==========================================
+ Coverage   65.30%   65.45%   +0.15%     
==========================================
  Files          16       16              
  Lines         709      715       +6     
  Branches      205      205              
==========================================
+ Hits          463      468       +5     
- Misses        165      166       +1     
  Partials       81       81              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.