Use math instead of KDTree for std::complex<float> -> AMP8I_PHS8I conversion

[JDanielSmith]

Copied from here:

---

I can add more detail, but be warned that my powerpoint skills aren't amazing and this is best drawn on a whiteboard or graph paper. The 8-bit amplitude and phase values that we're mapping to have a systematic distribution in the complex domain. Here's what I mean by systematic distribution:

A couple things to point out:

• The graph only shows 16 phase angles and 7 magnitudes per angle. A more realistic plot would have 256 phase angles and 256 magnitudes per angle.
• The phase_direction[] array contains a unit vector for each possible phase direction. Just the second vector is labeled in the plot.
• The phase_delta is the angular difference between each phase_direction.
• The magnitudes are the ordered set of possible magnitudes and they're identical for each phase angle.

Given any complex value V, we can determine the nearest phase angle by computing round(phase(V) / phase_delta). Determining the nearest magnitude value for V is slighly more complex:

The complex value V is projected onto the nearest phase_direction via the dot product. The resulting green point is then what's used to find the nearest magnitude via binary search (std::lower_bound).