Update maybe-batching logic.

1. Don't re-evaluate state for already running jobs. If the state
   changed, JobSchedulerService would be notified about it.
2. Only send a MSG_STOP_JOB message for jobs that are actually running.
   If the job isn't running, then just remove it from the pending list
   (if applicable).

Bug: 138469672
Bug: 141645789
Test: atest CtsJobSchedulerTestCases
Change-Id: Ia0d9f8845c81fe927193e63a0957f6a3ae0e202a