Pass Through chunks arg in write_zarr to layers

[ilan-gold]

Would it be possible to replicate this functionality:

https://github.com/theislab/anndata/blob/2723c9c9e8bc0fe50a9513f292916616a603923b/anndata/_io/zarr.py#L48-L51

in the layers as well?

https://github.com/theislab/anndata/blob/2723c9c9e8bc0fe50a9513f292916616a603923b/anndata/_io/zarr.py#L58

Maybe there is a reason not to do this though

[ivirshup]

So, I'm a little bit hesitant to define much about the behavior of chunks at the moment. This is because we'd like to integrate dask with anndata, where chunks need to be aligned between arrays that are used together. In this instance, it may make more sense if sparse arrays aren't chunked along the indices or data but the sparse arrays themselves would be chunked by their cartesian indices (e.g. you'd get chunks of sparse arrays instead of chunks of dense arrays).

I don't see any harm in adding the behavior you suggest right now, but it could change soon. I think there would be value in being able to specify these things separately.

[ilan-gold]

@ivirshup I agree that they should be specified separately, especially since they layers and X might be different types. Perhaps in order to wrap up #524 as well, we could just add a sparse_chunks argument to write_zarr? Then these could both be passed on to the layers?

[flying-sheep]

@ilan-gold is this issue still relevant?