feature request: embed cells using subset of genes

[rpeys]

I think it would be useful if a user could specify a list of genes (eg. highly variable genes) that they want to use as input to scgpt.tasks.cell_emb.embed_data() so that the function can run faster/exclude noisy genes without having to subset the anndata object, as the user may want to use the full, unsubsetted object for downstream analysis. This would be similar to the way that the scanpy.pp.pca() function works in scanpy.