Known issues shipped in v0.8.0 from the pre-release reviews

[xsmyile]

Pre-release reviews of v0.7.0..v0.8.0 found these issues and left them open for v0.8.0. None of them loses stored mail.

• restore-backup can't resume after an interruption between its two renames (crates/rustmail-storage/src/migrate/restore.rs). db is then missing and .schema0.bak exists, and both a new restore and a normal start refuse that state. The data survives; recovery is manual.
• A message can commit after SMTP has already answered 451 (crates/rustmail-server/src/main.rs store_group, and the timeout in crates/rustmail-smtp/src/session.rs). The session times out while insert_batch waits on the write lock, so the sender retries and the message is stored twice. The ack deadline needs to be coordinated with a write still in flight.
• A reconcile in place drops new results that fall between rows already loaded (ui/src/stores/messages.ts, reconcile). For example, a message starred during a disconnect, whose date lands inside the loaded window, never appears.
• Replay mistakes a snapshot row past the first page for a new arrival (ui/src/stores/messages.ts replay, crates/rustmail-tui/src/app.rs replay). This happens with more than one page of arrivals during one fetch, and the total ends up one too high.
• A swap window during migration (crates/rustmail-storage/src/migrate.rs, swap). A v0.7.0 process on the same file could commit and close all its connections between the source lock's release and the renames; its write would then stay only in .schema0.bak. This needs two instances on one file and a window of a few milliseconds. The fix is to re-fingerprint the backup after the renames and roll back if it changed.

[xsmyile]

• A message that starts matching a filter elsewhere doesn't enter the list (ui/src/stores/messages.ts patchMessage; the same applies to the TUI). Starring, tagging or marking a message unread from another tab while this one shows a filter doesn't bring the message into the list or the total. A refetch on unknown ids was tried in xsmyile/pre release fixes #127 and dropped: an unknown id is also a row the list let go after a flag change took it out of the filter, and the refetch then raced the in-flight PATCHes (the e2e unread filter › walks forward total came out wrong).