graph method for PDHMM event groups that unifies finding/merging and overlap/mutual exclusion

[davidbenjamin]

@jamesemery The fun begins. No change in output yet, but a non-trivial change in implementation.

Instead of making preliminary event groups according to overlap, then merging them according to mutually excluded events, this PR does it all in one step while automatically handling transitivity by treating as a matter of finding connected components of a graph whose vertices are events and whose edges are reasons (overlap and mutex) for events to be in the same event group.

All the failures are from WDL tests. I assume those are not related.

[codecov]

Codecov Report

Merging #8366 (d374080) into master (a86d8a3) will decrease coverage by 0.021%.
The diff coverage is 70.742%.

❗ Current head d374080 differs from pull request most recent head 62166d4. Consider uploading reports for the commit 62166d4 to get more accurate results

Additional details and impacted files

@@               Coverage Diff               @@
##              master     #8366       +/-   ##
===============================================
- Coverage     86.314%   86.293%   -0.021%     
- Complexity     39474     39480        +6     
===============================================
  Files           2358      2358               
  Lines         185234    185226        -8     
  Branches       20323     20339       +16     
===============================================
- Hits          159882    159837       -45     
- Misses         18196     18223       +27     
- Partials        7156      7166       +10     

Impacted Files	Coverage Δ
...walkers/haplotypecaller/HaplotypeCallerEngine.java	67.045% <0.000%> (ø)
...tools/walkers/haplotypecaller/ramps/RampUtils.java	2.198% <0.000%> (ø)
...ava/org/broadinstitute/hellbender/utils/Utils.java	80.639% <33.333%> (ø)
...pecaller/readthreading/ReadThreadingAssembler.java	58.481% <41.667%> (-4.244%)	⬇️
...alkers/haplotypecaller/AlleleLikelihoodWriter.java	81.429% <66.667%> (ø)
...PartiallyDeterminedHaplotypeComputationEngine.java	76.123% <73.869%> (-3.587%)	⬇️
...institute/hellbender/utils/haplotype/EventMap.java	73.636% <100.000%> (ø)
...nstitute/hellbender/utils/haplotype/Haplotype.java	91.346% <100.000%> (ø)
.../utils/haplotype/PartiallyDeterminedHaplotype.java	88.136% <100.000%> (ø)
.../hellbender/utils/haplotype/HaplotypeUnitTest.java	97.561% <100.000%> (ø)

... and 2 files with indirect coverage changes

[jamesemery]

Neat. It certainly reads better... the jury is still out on wether the graph overhead and the graph connectivity inspector is necessarily the most efficient approach here when there are a lot of events... I doubt this is going to be that severe cost however. I also think this is going to be much easier to extend for the jointdetection code than the current factoring which is good...

I would consider running a much larger section (like a chromosome) to inspect for mismatches with the old code... i'm not sure i'm convinced that the integration is sufficient to catch edge cases with this code

[davidbenjamin]

@jamesemery Ready for re-review.

I ran this branch against master in --dragen-378-concordance-mode on a PCR plus NA12878 sample over all of chromosome 20. The output was completely identical and runtime appeared unaffected. If anything I think this branch had a tendency to be faster on all of the longest-running of the 100-way scattered jobs.

[jamesemery]

Thank you for double checking the outputs on a larger section. 👍