Umi-tools

- Adds options to specify a delimiter for a cell barcode or UMI which should be concatenated + options to specify a string splitting the cell barcode or UMI into multiple parts, of which only the first will be used. Note, this options will only work if the barcodes are contained in the BAM tag - if they were appended to the read name using `umi_tools extract` there is no need for these options. See 217 for motivation:
- `--umi-tag-delimiter=[STRING]` = remove the delimeter STRING from the UMI. Defaults to `None`
- `--umi-tag-split=[STRING]` = split UMI by STRING and take only the first portion. Defaults to `None`
- `--cell-tag-delimiter=[STRING]` = remove the delimeter STRING from the cell barcode. Defaults to `None`
- `--cell-tag-split=[STRING]` = split cell barcode by STRING and take only the first portion. Defaults to `-` to deal with 10X GEMs
- Reduced memory requirements for `count --wide-format-cell-counts`: 222
- Debugs issues with `--bc-pattern2`: 201, 221
- Updates documentation: 204, 210, 211 - Thanks kohlkopf, hy09 & cbrueffer

Minor update. Improves detection of duplicate reads with paired end reads, reduces run time with `dedup --output-stats` and a few simple debugs.

* Improved identification of duplicate reads from paired end reads - will now use the position of the FIRST splice junction in the read (in reference coords) (187)
* Speeds up `dedup` when running with `--output-stats` - (184)
* Fixes bugs:
* `whitelist --set-cell-number --plot-prefix` -> unwanted error
* `dedup` gave non-informative error when input contains zero valid reads/read pairs. Now raises a warning but exits with status 0 (190, 195)
* `count` errored if gene identifier contained a ":" (198)
* Renames `--whole-contig` option to `--buffer-whole-contig` to avoid confusion with `per-contig` option. `--whole-contig` option will still work but will not be visible in documentation (196)

Version 0.5.0 introduces new commands to support single-cell RNA-Seq and reduces run-time. The underlying methods have not changed hence the minor release number uptick.

UMI-tools goes single cell

New commands for single cell RNA-Seq (scRNA-Seq):

* **whitelist** - Extract cell barcodes (CB) from droplet-based scRNA-Seq fastqs and estimate the number of "true"
CBs. Outputs a flatfile listing the true cell barcodes and 'error' barcodes within a set distance. See 97 for a motivating example. Thanks to Hoohm for input and patience in testing. Thanks to k3yavi for input in discussions about implementing a 'knee' method.

* **count** - Count the number of reads per cell per gene after de-duplication. This tool uses the same underlying methods as `group` and `dedup` and acts to simplify scRNA-Seq read-counting with `umi_tools`. See 114, 131

* **count_tab** - As per `count` but works from a flatfile input from e.g `featureCounts` - See 44, 121, 125

In the process of creating these commands, the options for dealing with UMIs on a "per-gene" basis have been re-jigged to make their purpose clearer. See e.g 127 for a motvating example.

To perform `group`, `dedup` or `count` on a per-gene, basis, the `--per-gene` option should be provided. This must be combined with either `--gene-tag` if the BAM contains gene assignments in a tag, or `--per-contig` if the reads have been aligned to a transcriptome. In the later case, if the reads have been aligned to a transcriptome where each contig is a transcript, the option `--gene-transcript-map` can be used to operate at the gene level. These options are standardised across all tools such that one can easily change e.g a `count` command into a `dedup` command.

Updated options:

* **extract** - Can now accept regex patterns to describe UMI +/- CB encoding in read(s). See `--extract-method=regex` option.

We have written a guide for how to [use UMI-tools for scRNA-Seq analysis](https://github.com/CGATOxford/UMI-tools/blob/master/doc/Single_cell_tutorial.md) including estimation of the number of true CBs, flexible extraction of cell barcodes and UMIs and per-cell read-counting as well as common workflow variations.

Reduced run-time (156)
Introduced a hashing step to limit the scope of the edit-distance comparisons required to build the networks. Big thanks to mparker2 for this!

Simplified installation ( 145 )
Previously extensions were cythonized and compiled on the fly using 'pyximport, requiring users to have access to the install directory the first time the extension was required. Now the cythonized extension is provided, and is compiled at install-time.

- Tweaks the way _group_ handles paired end BAMs. To simplify the process and ensure all reads are written out, the paired end read (read 2) is now outputted without a group or UMI tag. (115).
- Introduces the `--skip-tags-regex` option to enable users to skip descriptive gene tags, such as "Unassigned" when using the `--gene-tag` option. See 108.
- Bugfixes:
- If the --transcript-gene-map included transcripts not observed in the BAM, this caused an error when trying to retrieve reads aligned to the transcript. This has been resolved. See 109
- Allow output to zipped file with extract using python 3 104
- Improved test coverage (`--chrom` and `--gene-tag` options). Thanks MarinusVL for kindly sharing a BAM with gene tags.

Improves run time for large networks (see 94, 31).

Thanks to gpratt for identifying the issue and implementing the solution

When using the directional method with the group command, the 'top' UMI within each group was not always the most abundant (see comments in 96). This has now been resolved