[Feature] Compression via Nvcomp

[thomcom]

Hi! We've been talking about adding nvcomp to kvikio. I'm looking to add python bindings for Snappy, Cascaded, and Lz4 algorithms from nvcomp. In order to do so, we'll need to add the python bindings nvcomp.pyx and nvcomp.pxd to kvikio/python/kvikio/_lib and a wrapper for them, nvcomp.py. Once this is done I'll write tests.

A CMakeFlag -DUSE_NVCOMP=True will be added, disabled by default.

We're planning on using the nvcomp headers that are installed by cudf, which can be installed via conda, right?

I'm also looking into adding kvikio as another library option for https://github.com/trxcllnt/rapids-compose, which will make maintenance and development quite easy.

[thomcom]

Looking at cuDFs means for installing nvcomp, it installs it locally only as a cudf build dependency. I'm not sure how to proceed, either kvikio needs to know about cuDF's internal workings or it'll need more explicit dependencies, I think. I'm checking if the conda install of cudf has nvcomp.h etc, I doubt it.

[thomcom]

@madsbk I did find the nvcomp.h headers in the cudf conda build after burning most of the day trying to run side-by-side rapids-compose builds, one for kvikio development, but I had to abandon that route with little fruit.

[madsbk]

Implemented in #24