Ragged Array Reading is Slow(er than it should be)

[CSSFrancis]

Describe the bug

There appears to be a bug somewhere in zarr (and I would imagine h5py based on the fact that the loading speed is similar) with regards to loading ragged arrays.

To Reproduce

Steps to reproduce the behavior:

data = np.array([np.random.randint(0, 100, size=np.random.randint(0, 20)).astype(np.float64)
                  for i in range(i)], dtype=object)
s = hs.signals.BaseSignal(data)
s.save("data.zspy", overwrite=True)
%prun hs.load("data.zspy")

Expected behavior

Ideally some chunk would be compressed once and then loaded once on each save/load cycle. What is actually happening is different for .zspy and .hspy.

For .hspy each index in the ragged array is compressed individually and then uncompressed individually. This isn't efficient at all but isn't the worst case scenario.

For .zspy multiple indexes in the ragged array are compressed at one time but it seems like only 1 index is being uncompressed at a time. The result is that as you increase the number of indexes compressed (n) the time to uncompress is multiplied by (n).

Python environement:

• RosettaSciIO version: 0.0.2dec
• Python version: 3.9

Additional context

See #164 for more context

[ericpre]

Done in #169.