Improve Scramble accuracy for BWA and Dragen 3.7.8

[mwalker174]

Updates methods for Scramble calling. This includes several critical components:

1. Filtering of calls within reference MEIs overlapping indels and small deletion SVs. This was a major source of false positives for Scramble with WGS. The filter identifies such cases from the BAM/CRAM and Manta calls.
2. Improved SVLEN estimation. The previous method was overly simplistic, but length estimation now accounts for the orientation of soft clips and alignment strand. Note that this method's accuracy is greatly limited in cases where split reads are found on only one side of the breakpoint.
3. Deduplication. Scramble seems to emit duplicated calls in many cases. Now calls within a short window are collapsed appropriately.
4. Dragen 3.7.8 realignment. An apparent regression was introduced somewhere after Dragen 3.4.12 and after or as of 3.7.8 where small indels cause erroneous soft-clipping. This caused a massive number of false positive calls with Scramble. GatherSampleEvidence now automatically detects if Dragen 3.7.8 is used using the CRAM header (or if the is_dragen_3_7_8 flag is set explicitly). If so, it performs realignment of all soft-clipped reads near Scramble calls using BWA and re-runs Scramble off the realignment. The cost of this is very small and eliminates this artifact.
5. Updates the overlap breakpoint filter in ResolveComplexVariants to prioritize Manta over Scramble. A major source of Scramble false positives was at deletion sites. This filter catches such cases missed in the raw call filter from part (1).
6. SVA precision was initially low even after implementation of the above filters. A hard filter has been added to ApplyManualVariantFilter to remove Scramble-only SVAs flagged as HIGH_SR_BACKGROUND. This raised SVA precision from ~0.65 to ~0.85 in our tests, with a tolerably small loss of sensitivity. In the future, improvements and/or retraining of the GQ Recalibrator model may obviate the need for this hard filter.
7. Filter secondary/supplementary reads from consideration in the Scramble tool. This was bringing in further false positives. The Dockerfile has been updated.

This PR also includes changes to the README to recommend Scramble for MEI calling and deprecate MELT support.

These changes were extensively tested and validated on CRAMS aligned with both BWA and Dragen-3.7.8, including a full run of 1000 Genomes Phase 3 on 3.7.8.