Use numcodecs VLenArray for encoding variable length dataset chunks

[jreadey]

Doc on VLenArray is here: https://numcodecs.readthedocs.io/en/stable/vlen.html#vlenarray.
I'm pretty sure that this will be more performant than the run length encoding that is done in hsds/util/arraydata.py currently.
Care will be needed to ensure that data written in the original schema is still readable.