lemur function force conversion to dense matrix

Hello!

Thanks for the interesting package!

I'm testing it on one of our datasets, which contains about 1M cells and 1.5k individuals. When calling the `lemur` function, I noticed that the tool converts the sparse count matrix into a dense array. This occupies a huge amount of memory and thus limits the possibility of applying the method to large-scale data.

Briefly, I have a SingleCellExperiment object that I generated from a dgCMatrix, and data tables like this.

```R
sce <- SingleCellExperiment(
assays = list(logcounts = sparse_matrix),
colData = cell_info,
rowData = gene_info
)
```

Then I'm trying to run lemur as suggested in the tutorial

```R
fit <- lemur(sce, design = ~ phenotype + Exp,
n_embedding = 20, test_fraction = 0.5)
```

And I get this message:

```
Warning: sparse->dense coercion: allocating vector of size 172.0 GiB
```

The memory occupancy then quickly goes up to about 250 GB.

Am I doing something wrong? Is there a way to make the method work with a sparse matrix (the standard dgCMatrix used by SingleCellExperiment)?

Thanks!