fix(linux-sqlite): tell a locked file apart from a full server

[datlechin]

Why

A save into a SQLite file another program holds locked said "The server has no room for another connection right now". SQLite's SQLITE_BUSY mapped to the same DriverError::Busy the server engines use for a full server or pool. The SQLite doc also said lock contention shows only at BEGIN IMMEDIATE, but in a rollback-journal file a reader elsewhere makes the COMMIT fail too.

Change

• DriverError::Busy carries a BusyReason: NoFreeConnection for the server engines and the pool wait, FileLocked for SQLITE_BUSY. The app words FileLocked as another program or an editor tab with an open transaction holding the file's lock. The other engines keep their message.
• SQLITE_LOCKED is now the server's own error: without a shared cache it is a conflict inside one connection.
• The SQLite doc says where a lock shows (BEGIN IMMEDIATE for a writer, COMMIT for a reader in rollback-journal mode, never for a reader in WAL), and release criterion 5 moves to Holds today.

Verified

file_locked.rs holds a real lock from a second connection and checks a structure change, a commit held up by a reader, an import, an editor run, reads and opening the file: each fails as Busy(FileLocked) and the file keeps none of the save. write_batch_takes_write_lock_before_probe and session_read_concurrent_with_workspace_write_busy_or_commit_never_hangs now expect FileLocked.

Not in this PR

• SQLite's pool wait (the one workspace connection held past 5 s) still reads "The server has no room for another connection".