Improve RD testing to recover overlapped CNVs > 100 kbp

[kjaisingh]

Description

This PR includes code to improve the depth testing methods in order to capture large CNVs that are overlapped by another CNV of a different SV type but called as two distinct events.

During our DRAGEN-SV benchmarking evaluations, we noticed 3 real CNVs > 100 kbp that were dropped following FilterBatchSites. Upon closer inspection of their characteristics, we came to the conclusion that:

• All 3 of these CNVs were overlapped by another distinct CNV of a different SV type, but had far fewer samples called for it relative to the overlapping CNV.
• The background depth levels for the dropped CNV were hence distorted.
• This resulted in the RD_log_pval values for these CNVs being shockingly low, despite the CNVs clearly appearing to be real events.

This PR introduces the following enhancements to ensure that such CNVs are not dropped:

• The singlesampZ test now uses a robust z-score based on median depths rather than means, which is distorted when there is an overlapping event of a different SV type in the background.
• twosampPerm test: TBD.

Testing

The following deck has testing results so far.

Pre-Merge Changes Required

• Conduct end-to-end testing.
• Copy over changes into RdTestV2.R.