Implement FDK filtering on GPU

[AnderBiguri]

Suprising I don't have an issue for this:

FDK filtering is slow now because its done in the CPU (and does some ugly transposes).
We should just accelerate this, it makes no sense to have such a slow version of FDK.

Note: Not all FFT libraries in GPU/CPU are compatible with BSD license.

[tsadakane]

Use cuFFT?

[AnderBiguri]

Sounds good actually.

[AnderBiguri]

I'll close this for now, as #423 suggests this is not a good idea.