v0.2.0-alpha.2

spdf 0.2.0-alpha.2

Second pre-release. See CHANGELOG.md. This is the first release cut
through the automated workflow in .github/workflows/release.yml and
the first with correct MIT license metadata on crates.io.

spdf 0.2.0-alpha.1

spdf 0.2.0-alpha.1

First pre-release on crates.io. Not stable. The public API surface
can change until 0.2.0 proper. Track the full list of changes in
CHANGELOG.md.

Highlights

• 9-fixture benchmark corpus (IRS 1040, IRS W-9, NIST SP 800-63B,
  NIST SP 800-53r5, RFC 8446, RFC 9110, Unicode CJK chart, encrypted,
  malformed).
• Spatial-precision benchmark in addition to token accuracy: spdf
  0.679 mean IoU vs liteparse 0.482, token F1 80.6 % vs 77.4 %,
  wall-clock 366 ms vs 2594 ms.
• Resource guards (timeout_secs, max_input_bytes, max_pages)
  with integration tests.
• Fuzz harness (cargo +nightly fuzz run parse_pdf). One known OOM
  mitigated at the fuzz target level — see fuzz/README.md.
• Windows, macOS, and Linux CI matrix (OCR on Linux + macOS only).
• New CLI flag --preserve-small-text for form-heavy PDFs.

Prebuilt binaries

• x86_64-unknown-linux-gnu — attached (self-contained; bundled
  libpdfium; Tesseract OCR not compiled in — use
  --ocr-server-url for HTTP OCR).
• x86_64-apple-darwin — build on macOS Intel.
• aarch64-apple-darwin — build on Apple Silicon.
• x86_64-pc-windows-msvc — build on Windows.

To add a binary, follow the recipe in the
README install section.

Verify

sha256sum -c spdf-0.2.0-alpha.1-x86_64-unknown-linux-gnu.tar.gz.sha256