Fixed an optimization in KBestHaplotypeFinder that caused some bestPaths to be lost while still preserving some of the optimization.

[jamesemery]

@davidbenjamin Was familiarizing myself with KBestHaplotypeFinder and decided to take a crack at this issue.

I have no idea how much of a performance hit this will end up being at extreme sites. At worst it involves adding more paths into the priority queues than existed before which could slow down the whole search algorithm.

Fixes #5907

[codecov]

Codecov Report

Merging #5952 into master will increase coverage by 0.002%.
The diff coverage is 100%.

@@               Coverage Diff               @@
##              master     #5952       +/-   ##
===============================================
+ Coverage     86.821%   86.823%   +0.002%     
- Complexity     32349     32356        +7     
===============================================
  Files           1993      1993               
  Lines         149474    149499       +25     
  Branches       16521     16521               
===============================================
+ Hits          129775    129799       +24     
- Misses         13675     13676        +1     
  Partials        6024      6024

Impacted Files	Coverage Δ	Complexity Δ
...s/haplotypecaller/graphs/KBestHaplotypeFinder.java	95.455% <100%> (ø)	23 <0> (ø)	⬇️
...ypecaller/graphs/KBestHaplotypeFinderUnitTest.java	98.547% <100%> (+0.094%)	81 <8> (+8)	⬆️
...nder/utils/runtime/StreamingProcessController.java	67.773% <0%> (-0.474%)	33% <0%> (-1%)

[davidbenjamin]

Nice work @jamesemery! The effect on performance is going to be negligible, and the effect on correctness is obviously beneficial. 👍

[mehrzads]

Thank you @jamesemery! It looks great.