Misleading exception logged finished/canceled job restarted once

In FLINK-24015, code was introduced to ensure exceptions are logged on the JM (Dispatcher) so we have them at least once in case they arise from the initialization.

To this end the code checks whether the failureInfo is set in the ExecutionGraph. On the face of it this makes sense, but this failureInfo is also set when a job is restarted, and never reset.

As a result we may print the exception that caused the last restart when the job is finished/canceled.