Improvement: Reassigned method mathematics

[Tremeschin]

I somewhat understand the math for this better spectrogram visualizations method, and have got it working at least once. However, I couldn't make it "fast"

(Note for future self, technical reasons)

• The interpolation of "target" bins for the shifted frequencies array of the FFT is expensive even for nearest(), is band pass the right way to go?
• How to reassign in time, what does that even mean? With frequency it's understandable, but I didn't see a relationship with temporal stability and "promises"; in that case should we write to a future pixel of the texture for all points?

Writing that just gave me an very interesting idea: what if we simply integrate all points we have in the FFT in the final texture target!?!?

No need to interpolate at all! I'll HAVE* to investigate this.

[Tremeschin]

For reference, this wikipedia subsection worked nice: https://en.wikipedia.org/wiki/Reassignment_method#Efficient_computation_of_reassigned_times_and_frequencies