general improvements

duphold generally now uses median instead of mean. this makes it less susceptible to outliers and less sensitive to variant calls that are not exact.

the cutoffs for DHD are now data-driven.

v0.0.7

• remove DHBZ for the z-score. this was less useful than the fold-change values.
• greatly improve DHD. the cutoffs for this are now data-driven so that we expect a low false-positive rate
  of ends that are called to have a rapid change in depth. this requires an extra pass over each chromosome that
  will add about 90 seconds of runtime per (human or similar-sized) genome.
• DHBFC is now based on the median instead of mean so it is less susceptible to outliers and off-center calls
• fix bug that in rare cases resulted in exiting with "how"
• also report DHSP for number of supporting read-pairs for the event. This is usually redundant with SVTyper info, but might
  be more accurate in some cases.