numerical stability improvements, add prior precision to posterior gaussian(s), minibatches for validation loss, test updates, bug fix

[dgreenberg]

add prior precision to all posterior components for Gauss/MoG proposals also add identity times Ptol, with Ptol defaulting to machine epsilon for the relevant data type use cholesky factorization to get log determinants of psd matrices new tests of symbolic math operations fix numerical conditioning in test of APT with Gaussian/MoG proposals reuse prior samples when testing APT with Gaussian/MoG proposals whitespace, better print statements, cleanup fixed bug where prior mean/covariance weren't normalized when prior_norm=True for APT with Gauss or MoG proposals *use minibatches for validation loss, and use the validation loss in a test

[dgreenberg]

So I think this version is ready to merge. If anyone wants to try it on their problem first just drop a comment here.