Compressed datasets can be very slow to fetch

I don't know if it comes from h5grove, h5web or a combination of the two.

This needs further investigation. It can be probably reproduced outside of Hibou by using a dataset with the same metadata:

Dataset
Type	Integer (unsigned), 32-bit, little-endian
Shape	2000 x 2167 x 2070 = 8971380000
Chunk shape	1 x 2167 x 2070 = 4485690

Compression filters
32008	bitshuffle; see https://github.com/kiyo-masui/bitshuffle

Also observed in jupyterlab-h5web