Reduce SVConcordance memory footprint

[mwalker174]

The SVConcordance tool is currently too inefficient in terms of memory usage, requiring several 100's of GB of heap space on ~100K samples. This PR aims to reduce memory usage in two ways:

1. Truth VCF records are stripped of all genotype fields except GT and CN, which are necessary and sufficient for concordance computations.
2. A new option --do-not-sort is introduced to skip output record sorting. A major source of heap usage is the output buffer in the ClosestSVFinder class, which ensures records are emitted in coordinate-sorted order. This buffer quickly fills, however, when there is at least one record being actively clustered that spans a large interval because the buffer cannot be flushed until a variant beyond the maximal clusterable coordinate of that large variant is encountered. This option will allow users to substantially reduce max heap usage on larger call sets (a single SVRecord can consume ~100MB with 100K samples).

Includes an integration test to cover the --do-not-sort functionality.

[cwhelan]

This looks fine to me -- it took me a little bit to re-understand the relationship between the "flush" methods in ClosestSVFinder and SVConcordance but it looks good to me.