Remove abandoned filter project and unneeded build dependency

[TedBrookings]

I'm never revisiting this effort (even if we somehow go back to the spark SV tool, I think there are better ways to approach this), so I wanted to get the xgboost predictor dependency out of GATK.

[gatk-bot]

Github actions tests reported job failures from actions build 2693262585
Failures in the following jobs:

Test Type	JDK	Job ID	Logs
integration	11	2693262585.12	logs
integration	8	2693262585.0	logs

[codecov]

Codecov Report

Merging #7950 (17c7434) into master (804d1e2) will decrease coverage by 0.009%.
The diff coverage is 100.000%.

@@               Coverage Diff               @@
##              master     #7950       +/-   ##
===============================================
- Coverage     87.078%   87.070%   -0.009%     
+ Complexity     36997     36905       -92     
===============================================
  Files           2217      2213        -4     
  Lines         173849    173276      -573     
  Branches       18791     18700       -91     
===============================================
- Hits          151385    150871      -514     
- Misses         15832     15833        +1     
+ Partials        6632      6572       -60     

Impacted Files	Coverage Δ
...tructuralVariationDiscoveryArgumentCollection.java	94.595% <ø> (-0.142%)	⬇️
...on/FindBreakpointEvidenceSparkIntegrationTest.java	96.296% <ø> (-3.704%)	⬇️
.../sv/integration/SVIntegrationTestDataProvider.java	93.333% <ø> (-0.784%)	⬇️
...spark/sv/evidence/FindBreakpointEvidenceSpark.java	69.710% <100.000%> (+0.225%)	⬆️
...er/tools/spark/sv/evidence/BreakpointEvidence.java	66.822% <0.000%> (-11.916%)	⬇️
...rg/broadinstitute/hellbender/utils/SVInterval.java	90.000% <0.000%> (-2.857%)	⬇️
...ute/hellbender/tools/spark/utils/IntHistogram.java	86.335% <0.000%> (-2.484%)	⬇️
...lbender/tools/sv/cluster/CanonicalSVCollapser.java	87.471% <0.000%> (-0.464%)	⬇️
...kers/haplotypecaller/AssemblyBasedCallerUtils.java	90.517% <0.000%> (-0.041%)	⬇️
... and 18 more

[cwhelan]

Yeah, fine to remove this stuff if we're not going to use it again. Nice to streamline dependencies and unused code out of GATK -- writing it was a learning experience but I think we've moved on to better approaches. My one question is -- aren't we going to add the xgboost dependency back in #7705 if that gets merged?

[TedBrookings]

No, this dependency is not from the main xgboost project, it's a 3rd-party library that allows you to make predictions using saved model files. At the time, the plan was to train models in python, then use this java tool to apply the trained predictor. The GQ filtering project trains using a java tool, so I have to bring in the real xgboost library.

[cwhelan]

Ah, so the ml.dmlc dependency is the real xgboost library and the tool for predicting from trained models? Got it, thanks for the clarification.