Truncate 128-bit floats as early as possible
postscript_utils.py
creates numpy.float128
arrays in two places: [1] [2]
We should:
- Confirm that these are still necessary: [1] is already subtracting off the max, so it may be possible to remove there.
- Rearrange the calculations such that they aren't necessary, if possible
- Truncate the numbers using
.astype(np.float64)
at the earliest point that doesn't measurably reduce our precision.