Methylprep

- mouse manifest updated to conform with illumina Genome Studio / sesame probe naming convention.
- mouse_probes.pkl now includes different probe types. Previously, if a probe type was 'mu' (multi)
or 'rp' (repeat) or IlmnID started with 'uk' (unknown?), it was moved to experimental mouse_probes.pkl.
This was about 6300 probes.
Now, all 'Multi' and 'Random' probes are moved and stored in mouse_probes.pkl, about 30,000.
- mouse manifest has a 'design' column with tons of human-readable notes on different probe origins,
including analogous EPIC human-mapped probes.

- pipeline CSV output will now include meth, unmeth, beta, and m-values for all probes, including failed probes.
version 1.4.0 to 1.4.5 was replacing these values with NaN if a probe was filtered by the quality_mask.
Pickled beta, M-value, noob_meth, noob_unmeth output files will continue to exclude (e.g. show NaN) probes that failed poobah_pval or quality_mask.

- fixed qualityMask for epic+

- faster circleci testing
- mouse probes have duplicate names, breaking dye-bias step, so it will fallback to linear-dye when duplicates are present
- added more mouse array test coverage

- now uses sesame's infer_type_I_channel function to detect and correct probe switching, if sesame=True
- uses sesame's nonlinear dye bias correction function, if sesame=True
instead of the previous linear-dye-correction in the NOOB function.
- as part of the run_pipeline(sesame=True) default ON settings, it will apply sesame's "quality_mask"
that automatically removes probes that are unreliable from all data.
- reads more IDAT raw data (run_info, probe nbeads, probe standard deviation)
- idat.py IdatDataset has new kwargs, including bit='float16' option to cut file/memory usage in half
by clipping max intensity at 32127 (which cuts off ~0.01% of probes)
- processing will mirror sesame more closely now, instead of minfi (to revert, use sesame=False in run_pipeline)
- adds sesame quality_mask, which auto-hides known set of sketchy probes.
- internal objects updated so that values align in every stage of processing
(i.e. if you apply the sesame quality mask, the output files and the SampleDataContainer will exclude those probes)
- make_pipeline provides a scikit-learn style interface, as alternative to run_pipeline

- ensures methylprep output matches sesame output
- order of probes in CSVs, pickles, and SampleDataContainer doesn't match
- fixes bug where are few probes had negative meth/unmeth values because of int16 limits.
Now it uses unsigned int16 data type and unit tests confirm no negative values appear.