Per-record checksum and corruption detection

[mairinkdev]

Summary

Add checksum (e.g. CRC-32) for each record (payload or header+payload—document which). Validate checksum on every read/scan; return a rich error (offset, file position, reason) when corruption is detected.

Scope

• Compute checksum when encoding (using existing header format from Day 2); verify when decoding in read/scan paths.
• Define checksum scope in docs/file-format.md (payload only vs header+payload).
• Error type: include offset, file position, and message (e.g. "checksum mismatch at offset X, position Y").
• Tests: append valid records; corrupt one byte in a record (in payload or checksum field); next read/scan must fail with the corruption error and not panic. Optionally: corrupt padding/magic and assert appropriate errors.

Acceptance criteria

• Every record read or scanned is checksum-verified.
• Corruption produces a deterministic, informative error (no unwrap in hot path).
• Tests prove that corruption is detected and reported correctly.

Suggested commit

feat: add per-record checksum verification and corruption errors