Vectorize model

[jvivian]

Model is built iteratively but may be faster with a vectorized implementation. The caveat is the $X_{d,g]$ priors will have to be informed from an observed X which will increase the total number of parameters.

[jvivian]

New model is a lot faster. For ~125 genes and 3 datasets:

• New model: 5m
• Old model: 19.5m

I'll need to run some comparisons between versions to ensure things are consistent before opening a PR.

[jvivian]

Had to figure out how to carry over the normalization term otherwise the laplacian error was too high to call things outliers. New implementation recapitulates old model's p-values, but is now much faster.