Batch the commit-to-disk of new persisted jobs

We aggressively commit persisted-job state changes to disk ASAP, but
this means that in some cases (both pathological and not) we wind up
redundantly writing the persisted-jobs description to disk multiple
times.  Now we reduce the potential write load by batching together all
updates within a given time threshold into a single commit to disk.

Bug: 125441169
Test: atest $ANDROID_BUILD_TOP/frameworks/base/services/tests/servicestests/src/com/android/server/job/JobStoreTest.java
Change-Id: If7b544064f35dc6b0543b2ec979ae94afb5d6e33