The shared memory counters fall back to using a regular malloc'ed segment when shared memory is not available. This doesn't seem useful, and means we do an atomic increment every time an operation does an update. So instead, if shared memory segments aren't available, the counters should do nothing.

Additionally, we don't have any tests that I'm aware of for the contents of the shared-memory counters. So if we introduce bugs in incrementing the number of active cursors or the number of commands executed, etc., we won't catch it. Create some basic tests that counters are correctly incremented and decremented in response to driver events.