Add support for GPU acceleration

[dsherry]

In the usability blitz, @christopherbunn showed what looked like an 8x speedup in wall-clock runtime when GPU support was enabled in our catboost component. So yes, GPUs are awesome :)

It appears sklearn has chosen not to prioritize GPU support. However, xgboost, lightgbm and catboost support GPU acceleration.

Proposed plan:

• Update catboost and xgboost components and pipelines to support turning on GPU acceleration as a boolean flag in the init, default false
• Once the automl strategy project #272 is in, we could write a GPU-specific automl strategy which only runs the pipelines which support GPU acceleration, with it enabled
• That means this is blocked on #272

Questions to consider:

• Should we find more GPU-enabled algorithms?
• Does this vary from platform to platform? We could choose to support only on linux initially.

[kmax12]

another library to look into that we could use to build components is cuML - https://github.com/rapidsai/cuml

[beckernick]

Happy to answer any questions about cuML, if folks are interested.

We've done a lot of work to tighten our interface compliance with scikit-learn since summer 2020, which has made it fairly smooth for AutoML tools like AutoGluon, TPOT, and PyCaret to add support for cuML.