Use cuda to improve speed for registers with many qubits

[aromanro]

The project is built with omp turned on and when it used the big matrices obtained by tensor product it benefited from it. Currently those big matrices are avoided, the tensor product matrix and its product with the state vector are optimized by taking advantage of the structure of the matrix.
The code for applying a quantum gate now does a single loop over the basis states. This could be parallelized using open mp, for a big register. I tried a quick hack but unfortunately for small registers there is a noticeable slowing down due of the omp overhead. I might try to avoid it in the future, for now the omp pragmas stay commented out there.

Nevertheless, for big registers cuda could be used (I would favor something that works on some other hardware, to not be tied to nvidia, such open cl or even compute shaders, but cuda is more performant and when performance is needed...), that could improve the speed quite a bit.