[MeownsterHunter]

Retail NSO files fail intake with a segment size mismatch. The parser reads the flags word from the wrong place in the header, so it thinks every segment is uncompressed. This PR reads the flags from the right offset and checks the segment hashes, so a wrong layout fails loudly instead of producing bad code.

What was wrong

The NSO header starts with the magic NSO0 at 0x0, the version at 0x4, a reserved word at 0x8, and the flags at 0xC (NSO format on switchbrew). crates/recomp-pipeline/src/homebrew/nso.rs read the flags from 0x8. Real files have zero there, so no segment looked compressed, and intake failed when the raw segment size did not match the decompressed size.

The test fixtures also wrote the flags at 0x8. The tests passed for that reason.

What this changes

1. Read NSO flags from header offset 0xC. The fixtures in crates/recomp-pipeline/tests/homebrew_intake.rs, crates/recomp-pipeline/tests/xci_intake.rs, and samples/homebrew-intake/generate.py now write the flags there too.
2. Verify NSO segment hashes when the header asks for it. Flag bits 3 to 5 request a check for the text, rodata, and data segments. The header stores the SHA-256 of each decompressed segment at 0xA0, 32 bytes per segment. The reader now compares them and returns an error on a mismatch.

Testing

• cargo test --workspace passes: 80 tests.
• I checked 8 NSO files from two retail games, 24 segments in total. With the flags read from 0xC, all 24 decompressed segments match the SHA-256 in their headers. With the flags read from 0x8, none match.