v1.1.0

Changed

• The implementation is split into private modules by concern (_retry, _connections, _stats, _sql, _errors, _database, _single_table). The public API is unchanged and from msqlite.msqlite import ... still works. BEGIN, single statements and COMMIT now go through one retry loop (LockRetrier) over a pure RetryPolicy; Transaction.execute, executemany and executescript share one timed cursor path.
• Log records go to the msqlite logger. They used to go to msqlite.msqlite, a child of it, so a handler or level on msqlite already received them; only a filter on the exact old name needs updating.

Fixed

• Lock-error detection assumed every sqlite3.OperationalError carries sqlite_errorcode. One constructed in Python rather than raised by the engine has no such attribute, so the documented message fallback raised AttributeError instead of running. Found by the new unit test for the fallback.
• pragmas=("PRAGMA",) (the prefix with nothing after it) was accepted and failed later when the connection opened; it is now rejected up front like an empty pragma.

Full changelog: https://github.com/jamesabel/msqlite/blob/main/CHANGELOG.md