spectrum: Precompute delay*doppler product and psd square root

Provides a 22% overall speedup (71% in the function) in a MIMO test scenario with 2 devices with 32 ports each. 😁

If you ask yourself why moving a multiplication outside a loop make that much difference, the answer is that these multiplications are complex ones, which translate to 4 double multiplications + 2 double additions. This is why the function is still slow.

Profiling results with -O0 -g

Before

After