🪲 BUG-#65: Store raw RFC822 source as bytes for binary-safe backup/restore

[FernandoCelmer]

Summary

Stop lossy UTF-8 round-tripping of RFC822 source. Backup/restore now preserves binary attachments and non-UTF8 bodies byte-for-byte.

Changes

• RawSerializer.file is now bytes (was str). A field_validator coerces incoming str via latin-1 (byte-preserving) for backward compatibility with any caller still passing strings
• RawModel.file switches Text → LargeBinary (BLOB) so SQLite stores the bytes verbatim
• Message.file follows suit (bytes), and Message.from_raw stores raw directly without .decode("utf-8", errors='replace')
• Sync.orchestrate writes entry.raw() instead of entry.text()
• MailBox.append receives bytes from Message.file unchanged (no more .encode("utf-8"))
• Tests cover binary round-trip across all 256 byte values, str-input coercion, and upsert with bytes

Migration

Existing rows stored as Text are still readable by SQLite; SQLAlchemy decodes the legacy column. New writes use the bytes path. If a consumer needs to migrate historical rows, re.encode("latin-1") on the stored string round-trips any byte sequence — but no automatic migration is shipped since the schema change is forward-compatible.

Closes #65

Test plan

• pytest — 211 passed (was 208 + 3 new bytes tests)
• bytes(range(256)) round-trips through save → get

[FernandoCelmer]

🔍 Code Review

Code issues found: 0

Correct end-to-end:

• RawSerializer.file → bytes with a latin-1 coercion validator (byte-preserving for the 0-255 range — the standard trick for opaque-bytes-as-str interop).
• RawModel.file → LargeBinary (SQLite stores as BLOB; legacy Text rows remain readable because SQLite is typeless).
• Sync.sync now writes entry.raw() (bytes) instead of entry.text() (lossy UTF-8 with errors='replace').
• Message.from_raw stores raw verbatim; MailBox.append no longer re-encodes a Message.
• Test asserting bytes(range(256)) round-trips through save → get is exactly the right invariant.

Note for release notes — this is a breaking API change for Message.file (str → bytes) and RawSerializer.file. Both are exported from email_profile. Current version is 1.0.0.dev1 so pre-release semantics apply, but the CHANGELOG should call this out before a stable 1.0.0 ships — any consumer printing msg.file will now see b'...'.

Closes #65.