Change to overlap-based method of amplicon resolution

[cjw85]

This PR changes how the amplicon (/primer pair) assigned to a read is determined in order to be:
a) a little more transparent
b) work more sensibly for fragmented amplicons
It also separates out the writing of the bam from the read selection to allow more sophisticated read selection post filtering.

Things proceed as follows,

First pass:

1. Determine amplicon by maximal overlap: assign this as the amplicon
   1b) Allow some filtering options based on nature of overlap
2. Determine if read overlaps primer sites for selected amplicon (and store a boolean for each)
3. Store information required for filtering

The reads are the downsampled per (amplicon, orientation), selecting reads preferentially on their span of the amplicon.

Second pass:

1. Trim reads according to primers
2. Write out reads and summary

[BioWilko]

p2_position is undefined, shouldn't this be:

print("ref end %s > primer_position %s" %(segment.reference_end, trim_end), file=sys.stderr)

[cjw85]

Yes, probably.