Documentation: Improve docs for zarr params

[felix0097]

Description of feature

We need to document better what the zarr config params are doing. Also provide some guidelines. E.g aim for a chunk size of 100Mb, n_obs_per_dataset that the total file still comfortably fits into memory (e.g 10-20GB) etc.

At the moment, this isn't really clear to the user I think

[felix0097]

this one is especially confusing:

[felix0097]

Also, the shuffle_chunk_size does not affect the chunk size of the underlying dask array:

annbatch/src/annbatch/io.py

Line 565 in c80f1ee

adata_concat = _lazy_load_anndatas(adata_paths, load_adata=load_adata)

In the past, we matched this. I guess this made sense as we loaded exactly those blocks during shuffling?