… subscribers

And instrument a bunch of actions.

Fix bug found for async execution mode when worker didn't correctly finish
shutting down.

Also, always instrument retry_all and release_many_blocked even if none is
retried/released, so we have that information as well, with the number of jobs
that actually were acted on.

And fix issue when discarding in batches, where the discarded count
was always being zero when the foreign key with cascade deletion is
present from jobs to ready executions. Discarding all jobs would delete
all ready executions in cascade, and then deleting them afterwards, for
the case where there aren't any foreign keys, would return 0 jobs
discarded.

This was sending two events per error instead of one.

And set log subscriber's logger to be Solid Queue's logger.

Also: refactor a bit how (OS) process attributes are exposed so they can
be logged more easily.

With Active Support notifications.

…vent

So we at least can tell when it received a SIGKILL or when it exited normally.