Use scikit-build for the build process

[vyasr]

This PR replaces a lot of the setuptools logic with using scikit-build, which allows us to leverage the existing CMake configuration for the C++ components of RMM to build the Python package using standard pip- and setup.py-based installs. Replaces #637.

[vyasr]

CC @kkraus14

[kkraus14]

One question, otherwise this looks amazing!

[vyasr]

rerun tests

[vyasr]

Ah good catch, I had noticed that earlier but forgot to address it. I think we'll need to bring this up with the skbuild team, I think skbuild automatically identifies subpackage by __init__.py even if we don't tell it to compile anything there. I agree that this isn't a blocker though.

[vyasr]

@ajschmidt8 and I just met and went through installing these conda packages into a fresh environment. Everything looks good, the package is well-formed and Python tests pass so at this point I think I'm comfortable proceeding. I think it's very unlikely that we will run into conda package issues at this point, and if someone finds bugs trying to build from source due to unexpected system configuration I expect them to be a small enough portion of the user base that we can work to resolve those issues if and when they arise.

[vyasr]

@robertmaynard @rapidsai/rmm-python-codeowners please take a final look at this PR, we can merge once you've approved.

[vyasr]

@gpucibot merge