"How to" example of direct GPU IO

[ivirshup]

Please describe your wishes and possible alternatives to achieve the desired result.

Create an example of loading from disk into GPU memory using kvikio and zarr.

[Intron7]

I'm currently testing kvikio with zarr and we need to very careful with the compression. Some compressions break kvikio. We can only use numcodecs.LZ4() or no compression. I would add a compression argument in write_zarr

[ivirshup]

I would also note that @Intron7 and I actually saw a performance decrease when using GPU with LZ4 compression. It seems like LZ4 decompression was actually quite slow at the moment, so we may want to wait on performance improving before creating this demo.

Having some trouble reproducing those at the moment though, due to some vscode issues.