[datlechin]

Why

• Release criterion 5 needs a test that kills the app during a save. Nothing checked what a SIGKILL mid-save leaves behind.

Change

• New harness-free test kill_during_save: TablePro runs as tablepro <file> in a child, saves a cell edit to a SQLite file, then saves two more. A trigger keeps the second UPDATE running, and the test sends SIGKILL with the journal on disk.
• The file must hold none of that save and take a write. The tablepro binary then restores the session, and the history, GSettings keyfile and saved tabs must read back with nothing moved aside.
• The session restore test's D-Bus helpers move to tests/support/app_bus.rs.

Verified

• Passes with CI's environment and the C locale. With the save changed to commit each UPDATE alone, it fails on a half-applied save.

Not in this PR

• Network engines: the server rolls back when the socket closes, and conformance covers a lost connection mid-write.