Audit Mutect2/HC FeatureCache access patterns

[cmnbroad]

I ran the full test suite using a branch that throws if a tool ever tries to query the FeatureCache using a query interval that is earlier than, but on the same contig as, the one currently cached. Several tests failed, including a few of the Mutect2/HC ones:

Mutect2IntegrationTest.testContaminationFilter
Mutect2IntegrationTest.testDreamTumorNormal
Mutect2IntegrationTest.testGivenAllelesMode
Mutect2IntegrationTest.testPon
Mutect2IntegrationTest.testTumorOnly
HaplotypeCallerIntegrationTest.testGenotypeGivenAllelesMode

The FeatureCache assumes that queries are always increasing along a contig; the failures in this branch indicate that the caller is attempting to back up and re-query territory that has already been cached and then trimmed.

I didn't track down all of these cases, but the general pattern appears to be that active region determination results in initial caching and trimming, and then the same/similar territory is traversed again during calling, resulting in cache misses. It happens pretty frequently when running M2 tests, at least for pon and germline resource inputs; we should investigate how much a better caching strategy would help performance. If it would, we'd need #4902 at a mimimum in order to use a alternate cache strategy.

[cmnbroad]

Closing in favor of #5895.