Avoid dense matrices

[stephens999]

gbcd-workflow/code/fit_cov_ebnmf.R

Line 17 in ab8aaf2

fit.cov <- flash_init(dat, var_type = 0) %>%

To avoid forming YY' should we not run flashier with data = list(u=Y, d=diag(ncol(Y)), v=Y)
If Y is sparse then this could be much faster?
(Maybe for the example we looked at they are not big enough for this to be an issue? But
some single cell datasets surely will be?)