Leafcutter

- `prepare_results.R` for `leafviz` received a significant re-write including
- Removed dependence on `bedtools`
- Friendlier CLI arguments
- New documentation on the doc website
- smart initialization got regularization (deals with co-linear covariates) and robustness to a rare divide-by-zero bug.

- Changed PSI and deltaPSI calculations to better match empirical and be consistent across runs (issue 41)

- Important bugfix for smart initialization when there are no technical/confounding covariates.

- LeafViz PCA now made with ratios rather than counts
- prepare_results now accepts covariates in support file
- Added magrittr and stringr dependencies
- LeafViz now allows colour and shape of points in PCA to be set to group or covariate

- Better control of Differential splicing initialization: the default "smart" initialization is now a one-shot method of moments estimator for the null model regression coefficients, but random initialization can be optionally specified, along with the seed (see --init and --seed options of leafcutter_ds.R) (issue 33)
- Dalila Pinto contributed a new script 'gtf2leafcutter.pl' for more robustly converting GTF files to the "annotation code" needed for leafviz. Thanks!

- `cluster_table` now includes `p.adjust` column calculated using BH FDR control.
- Improved gene view in LeafViz
- `check.names = F` when reading counts files to avoid issues with sample file names getting changed by R
- Removed `-w` option from shebang in perl scripts since this was causing issues on Linux (fine on Mac)