Add RAPIDS backend

[floscha]

In addition to pandas and Spark, add a third computational backend using cuDF and cuML. This way all processing can directly be performed on the GPU, hopefully resulting in significant performance gains.