Direct-IO fast paths + page-cache correctness and robustness fixes

[OBrezhniev]

Direct-IO fast paths + page-cache correctness and robustness fixes

Summary

Two performance fast paths for large transfers, followed by a series of
correctness and error-propagation fixes for the page cache, all covered by
new tests. snarkjs/binfileutils drive every zkey/ptau/wtns through this
library, so large sequential section reads/writes dominate its real-world
profile.

Performance

• Direct-read fast path: reads ≥ 1 MiB with no overlapping dirty pages
  copy straight from disk into the caller's buffer, skipping the page cache
  and its page→destination copy (the dominant cost of large sequential
  section reads).
• Direct-write fast path: same idea for writes ≥ 1 MiB to regions with no
  cached pages; any cached page in range falls back to the cached path so
  nothing goes stale.
• O(cached pages) range guards: the dirty/cached-page checks for the fast
  paths iterate the actually-cached pages instead of every page index in the
  range, keeping them cheap for huge ranges with small pages.

Correctness

• BigBuffer corruption fix: the direct paths hand the caller's buffer to
  fd.read/fd.write, which requires a real TypedArray/DataView. A
  BigBuffer (paged, not a view) silently corrupted data; both paths are now
  gated on ArrayBuffer.isView and BigBuffers take the cached path (which
  copies via their own .set()). Regression-tested with >1 GiB-styled
  section reads/writes.
• EOF destination-offset fix: a cached-path read past the end of a
  truncated file computed the destination offset from the EOF-clamped
  remaining count, silently shifting the valid bytes to the wrong position in
  the output buffer. Tracked independently now; the real bytes land at the
  start.
• Page-cache IO error propagation: a failed page read rejected only the
  first waiter — co-readers of the same page were queued on page.loading
  and awaited forever; the dead page also blocked retries. A failed
  background flush left page.writing set forever (pinning the page and
  wedging close()) and surfaced the error only at close() — which error
  paths often skip, silently truncating the file. Now: every waiter is
  rejected and the page dropped for retry; flush errors clear the flag, latch
  self.error, fail the next read/write fast, and reject close().

Compatibility

• Node detection at runtime (process.versions.node) instead of the
  webpack-only process.browser, which is undefined under Vite/esbuild/SES.
• browser field stubs fs/constants so bundlers need no hand-written
  stubs; open flags come from fs.constants rather than importing node:fs
  in the shared module.
• Dev-dependency audit findings fixed via overrides.

Validation

npm test: 21 passing, including the new regression tests for the BigBuffer
direct-io gate, the EOF offset shift, multi-waiter page-read failure, and
background-flush error surfacing. Each fix's test was verified to fail
against the pre-fix code.

Stacked work

The HTTP Range / Blob streaming backends build on this branch and are PR'd
separately from feature/url-blob-streaming.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Wo6AVSAwvL9mHTpvnREZPR