Plonk : reduce the number of ifft

IFFTs are no more computed at gate level : query’s precomputed_poly now carries evaluations that are kept until Polynomial_protocol.compute_T’s division. Thus only one IFFT is performed beside the IFFTs done to compute polynomials to be committed.

This allows less IFFTs but it may also imply bigger FFTs, especially for the wires at the begining, since all polynomials have to be evaluated on the larger domain reached. Therefore, the gain depends on the number of gates used & their degree.