* use dynamic scaling of the inputs to get accurate ratios (effectively
floating point) to maintain accuracy for small arguments
* this also allows shifting later and keep more bits
* use u32 ratio to keep one more bit
* merge the corner case unittests into the big test value list
* print rms, absolute and axis-relative angle
* simplify the correction expression to get rid of one multiplication
* use 5 bit for the correction constant and 15 bits for r
* least squares optimal correction constant, this lowers the max error
below 5e-5
* shrink the LUT by another bit
* correctly use the octant bit to offset the dphi to LUT entry midpoint
* add more diagnistics to the unittest and rewrite it in relative units
* MSB-align phase and output to match the PLL data, dynamic range and
remove the need for roudning bias.
* clean up the build.rs table generator a bit