The Joker v1.0

[adrn]

It's a little silly to even open a PR for this, since it's essentially a complete rewrite of the whole package, but opening it for myself to keep track of and to be able to point others to the branch for testing and discussion.

This improves a number of things relative to the "legacy" version of The Joker (what is currently in master):

• More flexible prior specification using pymc3 distributions
• A sensible default prior on the semi-amplitude, K, that keeps the prior on companion mass fixed (at fixed primary mass). Users have to decide how to set the scale of the prior at a reference period (which can be translated to a mass scale)
• Fixed a number of hanging issues related to random numbers
• Added support for generating multiple samples in the linear parameters at each nonlinear parameter sample that passes the rejection step (i.e. #62)
• Added support for processing the prior samples cache with randomly drawn batches from the iterative rejection sampler (i.e. #90)

TODO:

• [x] Enable support for calibration offsets - everything is in place to support this, just need to address some loose TODO items (i.e. #79)
• [x] Provide a bridge to exoplanet radial velocity fitting (i.e. #89)
• [x] Update the documentation / review API docs
• [x] Write so many tutorials

[coveralls]

Coverage decreased (-5.4%) to 82.94% when pulling 80c01b458534b1a7dac0bdc23005c53978696ad4 on thejokerv1-simpler into 132cb53498ce067bbf18d6788e3a2beb5dfd7cf3 on master.