Handle missing notifications in recv/get_event

[maxdml]

Today it is possible that a notification from the backend is missed and recv/get_event sleep until the entire timout, even as the awaited value is present in the database.

Consider this scenario:

• Postgres emit a notification on the connection opened with the application
• Postgres crashes before the notification is sent on the wire
• Postgres restart, the notification connection is re-established (by error handling on the application side), but never re-emit the notification
• The application is never notified.

In TypeScript, we handle this by periodically polling Postgres. In Golang, we have a mechanism for the listener loop to signal, through a channel, all the consumers that they should read from the database because the listener connection was lost.

Ideally we'd implement the later in Python.

[kraftp]

@chuck-dbos Another improvement we should probably bring over from TS