Blueice

* Prevent negative rates being passed to Barlow-Beeston equation, and allow per-event weights (32)
* Add likelihood that takes coupling as shape parameters (34)
* Patch for tests (37)
* Use scipy stats for PoissonLL (40)
* Do not scale mus when livetime_days is 0 (41)

* Likelihood sum wrapper (17)
* emcee bestfit and multicore precomputation (18)
* LogAncillaryLikelihood for constraint terms (19)
* HistogramPDFSource simulation, order shape parameter dict (20)
* Efficiency shape parameter, LogLikelihoodSum enhancements (23)
* Use scipy as default optimizer (24)
* Minuit support for bounds and errors (26, 27)
* Per-source efficiencies, weighted LogLikelihoodSum (28)
* Use atomicwrites for cache to prevent race conditions (30)

Several improvements, detailed description to follow

- Binned likelihoods (7)
- Argument validation for LogLikelihood function (8)
- Automatic handling of statistical uncertainty due to finite MC/calibration statistics (9):
- Adjustment of expected counts per bin using Beeston-Barlow method for one source
- Generalized to multiple sources, but only one with finite statistics.
- Only for binned likelihoods.
- iminuit integration, use as default minimizer if installed (10, 13)
- compute_pdf option to do full likelihood model computation on the fly (11)
- HistogramPDF to provide just histogram lookup/interpolation from DensityEstimatingSource (12)
- inference functions -> LogLikelihood methods
- Most-used functions/classes available under blueice (blueice.Source, blueice.UnbinnedLogLikelihood, ...)
- compute_pdf auto-called, consistent handling of events_per_day
- Start of documentation, readthedocs integration

- Big internal refactor, some API changes (5)
- DensityEstimatingSource
- Bugfixes, more tests

- Renamed to blueice, XENON stuff renamed to laidbax
- Experimental radial template morphing (4)
- Tests, several bugfixes (e.g. 3)
- Rate parameters are now rate multipliers
- Linear interpolation of density estimator
- Parallel model initialization