sqlite: fix error 'database is locked'

[elv-gilles]

The fix add retries for the case where multiple connections are used simultaneously and uses a mutex for multiple write calls made concurrently with a single SQLiteConnection (note that the code in the fabric uses a single connection for the server, the client and the inspector).

• use option _txlock=immediate when opening a read-write connection in db.DB. See:
  • IMMEDIATE in sqlite doc for transaction
  • _txlock in go-sqlite3 doc
  • this comment in #607 of go-sqlite3
• add lock around SQLiteConnection.Execute
• add retries for error sqlite_busy (when using multiple connections). see SQLITE_BUSY in sqlite doc
• improve rqlite errors by reporting the caller

---

sample error messages found in logs

2023-05-23T19:37:37.642Z DEBUG enqueue errors: rqlite.enqueueMessages 
  - rqlite error: unexpected result count: 1, expected: 2  -> database is locked "INSERT INTO ltaskq_asynq_queues (queue_name, state) VALUES(dist:livepart, 'active')  ON CONFLICT(queue_name) DO NOTHING;", state: pending, type: dist:livepart, id: hqt_CUZL75rJyNh2vXWrSGLemQeBaEaJXHpKr44rPvVT logger=/store/localTaskq gid=9435900

2023-05-23T19:39:42.915Z WARN  Could not move task id=hqt_9PXGZrwAkWCYDYdduxYq5tAEjXhRz8GUngcNShHD:inod21LrUiai85M8hq24yana53KNdY9i 
  type="dist:livenode" from "asynq:{dist:livenode}:active" to "asynq:{dist:livenode}:completed":  rqlite.setTaskCompleted: NOT_FOUND: error 
  (UPDATE ltaskq_asynq_tasks SET state=completed, task_msg=0a0d6...8f9acb4a306, deadline=NULL, sid=inodKnnpcZjRT65FLteMrULXBme4xS9, done_at=1684870777, cleanup_at=1692646777, retain_until=1687462777 WHERE task_uuid=hqt_9PXGZrwAkWCYDYdduxYq5tAEjXhRz8GUngcNShHD:inod21LrUiai85M8hq24yana53KNdY9i AND state='active'): 
  database is locked; Will retry syncing logger=/store/localTaskq gid=9438871

[jslching]

LGTM as far as I can tell