Parity: decode BMP headers from generated ExifTool tables

[thedancingdeveloper]

Summary

• replace the hand-written BMP field reader with the generated ExifTool BMP::Main and BMP::OS2 binary tables
• reproduce ExifTool 13.30's BMP-specific conversions, conditions, and 68-byte AVI header hook
• add focused tests for Windows V3, OS/2 V1, top-down height, FourCC compression, AVI layout, and rejected DIB lengths

Oracle and source

ExifTool 13.30 only:

• lib/Image/ExifTool/BMP.pm: %Image::ExifTool::BMP::Main
• lib/Image/ExifTool/BMP.pm: %Image::ExifTool::BMP::OS2
• lib/Image/ExifTool/BMP.pm: ProcessBMP
• corpus: all 190 files in t/images

Parity evidence

The full conformance classifier was run before at base 6fb64c5f246e8dde268c0597626a13ff3c2b9593 and after at 00f11751b545f92497d9bb60aa533c22f01434eb.

BMP	matched	rename	value difference	missing	extra
before	10	0	1	5	5
after	16	0	0	0	0

Corpus totals moved from matched 5,670 / rename 11 / value difference 885 / missing 3,123 / extra 1,349 to matched 5,676 / rename 11 / value difference 884 / missing 3,118 / extra 1,344. A per-format JSON diff confirms BMP is the only changed format.

The focused incremental comparator was also run as required, but its ExifTool extractor intentionally filters the entire File family; BMP's real header tags are in that family, so it reports 0 ExifTool tags and cannot measure this slice. The conformance classifier does compare those tags and is the authoritative evidence above.

Validation

• CARGO_BUILD_RUSTC_WRAPPER=sccache cargo test --lib parsers::image::bmp::tests -- --nocapture: 5 passed
• bash scripts/verify_parity_fast.sh --skip-corpus: passed
• CARGO_BUILD_RUSTC_WRAPPER=sccache cargo clippy --lib -- -D warnings: passed
• cargo fmt --all -- --check: passed
• git diff --check: passed
• generated tables/generators were not changed, so regeneration was not required

No AI harness, fleet, model fixer, model reviewer, or external model/API command was run.