Interpolated linear transformation on PSFs and Wavefronts

[h-greer]

Right now Wavefronts support scale_to and rotate as interpolation operations, and PSFs just support rotate. It might be better to support a single generalised linear interpolation that perhaps takes a CoordTransform as input, to allow for cases such as the WFC3 detectors where the existing transformations don't cut it.

[LouisDesdoigts]

Yep this has been something I've been thinking about for a while (see #290). Thankfully interpax exists now, and I've used it extensively through Amigo. For PSF interpolation its a no brainer (modulo maybe some minimal wrappers that prevent nonphysical PSF's, such as those with negative values).

Interpolating Wavefronts is a much more complex task. Using amplitude-phase (outside of a few cases) has issues as interpolating through phase wrapping can cause a large host of undesired and non-physical behaviors. similar issues happen on real-imaginary, but I did a small investigation with @ataras2 recently and found it could work in some specific cases (@ataras2 do you remember where we landed on this?).

Theres definitely some upgrades to happen here, mainly swapping to interpax in utils, and then providing a unified API for this across dLux, something akin to CoordTransform, though I'm a little unconvinced that is the best API/entirely necessary. Which is essentially a long way of saying yes lets do it, but there are a few other pieces to think about, i.e. #295, comments of #289.

Happy to discuss this more, let me know your thoughts.

[ataras2]

iirc tracking the phasor (i.e. real and imag) was best since derivatives at zero amplitude are still defined.
For interpolation, it worked fine for the case we were looking at, where we interpolated wf(x,y; optical_param) wrt. optical_param, and this was done with just real and imaginary components. I guess the condition here was that the samples were selected to only have small variations in phase - never changing more than pi.