Negative binomial null residuals

[willtownes]

As a user with data in a SingleCellExperiment, I can compute either Pearson or deviance negative binomial null residuals so that I can use them in downstream analysis.

• data in dense matrix in memory
• data in DelayedArray on disk
• (maybe) sparse Matrix data in memory

Possible implementation paths:

1. Wrap the sctransform CRAN package. This provides regularized negative binomial Pearson residuals. It doesn't appear to support disk-based data though.
2. Do our own implementation with glmGamPoi package as backend for fast nb regression.