endianness character present in .zattr for single byte data types

[grlee77]

cc @joshmoore

While running a branch of zarr_implementations with jzarr-based read/write tests (with jzarr 0.3.2), I encountered the following issue.

The data used in the example in that repository has an 8-bit unsigned integer format. The .zattr JSON files created by jzarr currently have the following entry

"dtype" : ">u1",

but .zattr files created by other libraries such as zarr-python, xtensor-zarr and z5py have:

"dtype" : "|u1",

It seems that the spec indicates that | should be used rather than > or < for data types with a single byte width where there is no endianness. A similar issue may also be present for |b1 and |i1 types, but I have not confirmed that.

[joshmoore]

Failing builds can be found in zarr-developers/zarr_implementations#37

[SabineEmbacher]

done