Umi-tools

Due to a bug in pysam.fetch() paired end files with a large number of contigs could take a long time to process (see 93). This has now been resolved.

Thanks to gpratt for spotting and resolving this.

**Added functionality:**

- ***Deduplicating on gene ids*** ( 44 for motivation):
The user can now _group_/_dedup_ according to the gene which the read aligns to. This is useful for single cell RNA-Seq methods such as e.g CEL-Seq where the position of the read on a transcript may be different for reads generated from the same initial molecule. The following options may be used define the gene_id for each read:
`--per-gene`
`--gene-transcript-map`
`--gene-tag`

- ***Working with BAM tags*** (73, 76, 89):
UMIs can now be extracted from the BAM tags and_group_ will add a tag to each read describing the read group and UMI. See following options for controlling this behaviour:
`--extract-umi-method`
`--umi-tag`
`--umi-group-tag`

- ***Ouput unmapped reads*** (78)
The _group_ command will now output unmapped reads if the `--output-unmapped` is supplied. These reads will not be assigned to any group.


\+ bug fixes for group command (67, 81) and updated documentation (77, 79 )

Improves the group command:
- Adds the --subset option as per the dedup command (74)
- Corrects the flatfile output from the dedup command (72)

The code has been tweaked to improve run-time.
See 69 for a discussion about the changes implemented.

- Corrects the edit distance comparison used to generate the network for the directional method.
- This will only affect results generated using the directional method and `--edit-distance-threshold` >1

Previously, using the directional method with the option `--edit-distance-threshold` set to > 1 did not return the expected set of de-duplicated reads. If you have used the directional method with a threshold >1, we recommend updating UMI-tools and re-running _dedup_.

- Debugs python 3 compatibility issues
- Adds python 3 tests