--ignoreDuplicates could use some improvements

[dpryan79]

@thomasmanke happened to notice that deepTools estimates for the number of duplicates are always significantly below those produced by picard. While there are a number of differences between the two algorithms (picard doesn't always get it right), one big difference is that deepTools only remembers the last alignment seen. That's not an issue if one uses picard to sort alignments, since they're then sorted by pos and then mpos, but that's not the case for samtools sort, which just sorts by pos. I already implemented an improved method here, but I'm not going to put that sort of change in a bug fix release, which is what the current develop branch will quickly become.

[dpryan79]

This is now implemented in the develop branch.