Implemented `side` parameter (resolves #1)

[baldassarreFe]

This PR includes:

• C++ code for searchsorted with correct support of side=[left|right]
• CUDA code for searchsorted with correct support of side=[left|right]
• helper functions for broadcasting numpy arrays and pytorch tensors to be used in searchsorted
• more unit tests to make sure all of the above work
• minor changes in the README (how to check CUDA availability during installation, how to run tests, how to run the benchmark)
• new benchmark script that uses the builtin timeit

Please check out the new implementation (the CUDA and C++ part should be much more readable), run the unit tests and let me know if you get similar times in the benchmark ;)

[aliutkus]

thank you very much I'll review this shortly

[baldassarreFe]

I've noticed the speed issue, I think it could be related to the left/right side, which at least in this version gives correct results. Anyway, I'll look into speeding up the implementation ;)