Fix possible deadlock on sql error on replay

When an exception was thrown while doing replay, there was a deadlock because one of data processors was waiting on a condition variable that was already destroyed by queries_commit_data_processor.
To fix this add a call to cancel in ~data_processor, which will make
sure that join is called. To make sure join is not called twice, a check
is added to join function, which makes it return early if join was
already called earlier.