GenericScanKernel for pycuda

[adityapb]

I am working on a project for which I need to use a pyopencl-like GenericScanKernel. I have added GenericScanKernel to my fork that is almost identical to the one in pyopencl. I was wondering if you would consider merging it here if I were to make a pull request.

Thank you.

[inducer]

Sure, I'm in principle in favor. I would however prefer if the two implementations could remain the same, so that each can benefits from improvements in the other. This can be made easier by an abstraction layer. Something of the sort exists as pyopencl._cluda and pycuda._cluda.