Some improvements

[ldenti]

• fix some bugs
  • we don't miss one SFS per cluster anymore (improvements on low coverage since we have way less SFS)
  • fix reallocation while smoothing (CIGAR and query sequence lengths differ for reads in smoothed.selective.bam #16)
• change some heuristics
  • rolling to fixed treshold in smoothing (this way we don't filter somewhat good alignments)
  • clusters are split based on length ratio and not delta
  • while chaining we also check for SVs distance (too distant SVs shouldn't be chained)
• change alignment matrix
• update dependencies