The GracefulShutdownInterrupt and generic error handlers were identical
except for one line. It's easier (if slightly slower?) to use an
explicit isinstance check than to maintain two different handlers.

This lets the worker be called with a timeout, with the timeout's
implementation details hidden from process_visit's clients.

Like GracefulShutdownInterrupt, TimeoutInterrupt is raised
asynchronously, may leave the program in an invalid state, and may be
caught prematurely.

The timeout uses an alarm and a signal handler to interrupt application
code at will. This implementation is fairly robust, but assumes that
nothing else (especially sleep functions) uses SIGALRM and has the
usual caveats of signal handlers raising exceptions.

While Logger.exception is a handy way of printing stack traces, any log
message can include them. Include the stack trace as part of critical
logs instead of in a separate message.