Fix concurrency in imagery store

[oldmud0]

Apparently, connection pools in SQLite are still too dumb to synchronize properly without triggering SQLITE_TIMEOUT, so I just decided to limit the connection pool to 1 to force mutual exclusion. However, I may have accidentally discovered a deadlock condition, given that the connection pool caused a timeout. I hope this is not the case.

Even with and without this fix, the tests for everything - not just the store - are failing once in a blue moon, which doesn't seem to be a good sign. What's worse is that the errors do not point to anything in particular - for instance, one error was about rolling back a transaction while a transaction was not underway. I checked the code path and there isn't a path where this could ever happen. Another error was a Z CAM E1 backend test saying that a dummy function was called twice when it should have been called three times. These are some very peculiar synchronization problems.

[bbridges]

Limiting the maximum write to just one works fine. Test flakiness should probably be looked into at a later date.