Implement float pexp_complex

In this MR I propose the SIMD complex function pexp_complex for float, which is the vectorized counterpart of std::exp(std::complex<float>), as previously described in issue #2635

Rather than creating a brand new sincos function, I slightly modified the current implementation of sincos_float. The idea is that the implementation of pexp_complex will remain the same after the implementation of sincos_double, as they will share the same interface. I may have performed too many comparisons for the IEEE edge cases, so I would appreciate it if you spot any inefficiencies.