Fixed a rare non-determinism in the AdaptiveChainPruner

[jamesemery]

It turns out the java PriorityQueue implementation is is undefined in its behavior when elements are equal. It turns out it was possible for there to be non-deterministic behavior for chains that had identical scores and the same starting vertex (but different paths in the graph. I think this comparator should be airtight now. @davidbenjamin let me know what you think.

[codecov]

Codecov Report

Merging #7851 (9aa31e4) into master (72684d0) will decrease coverage by 0.001%.
The diff coverage is 100.000%.

@@               Coverage Diff               @@
##              master     #7851       +/-   ##
===============================================
- Coverage     86.948%   86.947%   -0.001%     
  Complexity     36927     36927               
===============================================
  Files           2219      2219               
  Lines         173673    173674        +1     
  Branches       18755     18755               
===============================================
- Hits          151006    151005        -1     
+ Misses         16055     16054        -1     
- Partials        6612      6615        +3     

Impacted Files	Coverage Δ
...rs/haplotypecaller/graphs/AdaptiveChainPruner.java	97.368% <100.000%> (+0.035%)	⬆️
.../hellbender/utils/python/PythonUnitTestRunner.java	75.410% <0.000%> (-3.279%)	⬇️
...itute/hellbender/tools/LocalAssemblerUnitTest.java	92.448% <0.000%> (ø)

[jamesemery]

@davidbenjamin Can I get a review on this? I'm not quite sure how to go about testing this (since i think the only reason it appeared in the first place was because it was a very complicated site that was forcing some edges to be filtered to satisfy the variants per graph limit (but non-deterministically).

[davidbenjamin]

@jamesemery Looks good!