Loading each cell as a .npy generates a large number of files.

[moi-taiga]

omics_h5ad_saver.py saves 1 file per cell.
This can become problematic when using large datasets, e.g. 600,000 or more cells.
Perhaps there is an efficient way to integrate omics_h5ad_saver.py into avae to avoid this without requiring too much memory at runtime?