Phase 2D + 3D refactor

[talonchandler]

This PR makes the following changes to the "Phase2D" and "Phase3D" reconstructions methods:

• moves these code paths from waveorder_reconstructor into their own files in /waveorder/models/phase2D_3D.py and waveorder/models/phase3D_3D.py.
• refactors both of these reconstruction paths into the following methods: calculate_transfer_function, visualize_transfer_function (for napari viewing), apply_transfer_function (very simple simulation), and apply_inverse_transfer_function. These methods are in preparation for direct calls from the recOrder CLI.
• performs all calculations with torch
• performs all calculations in CZYX order
• adds tests to the key parts of optics, utils, and the new examples
• preserves the code paths for the PTI and Birefringence_recon (S1&S2 diffraction-aware reconstruction) in waveorder_reconstructor. I am now testing several example scripts to alert me to breaking changes.

I suggest starting your review by running the new examples scripts in /examples/PODT_Phase_simulation/. This will demonstrate a simplified form of what I expect on the recOrder side: the TF is calculated (and optionally viewed in napari), saved to a file, then used to reconstruct the data.

Next, I suggest you take a look at the (currently under-documented but hopefully much clearer than before) function signatures for calculate_transfer_function and apply_inverge_transfer_function in models/phase2D_3D.py and models/phase3D_3D.py. These function signatures have a nearly direct correspondence with recOrder's GUI parameters.

Finally, I'll request your thoughts on my models format that arose naturally during my refactor. I've temporarily named the three models after our assumption about the sample (phase2D,phase3D, planaranisoND) and the data we acquire (3D, polND). To my eye, this seems to be an extensible way to split up the code since different assumptions about the object and/or different data will lead to different models (and code). I'm also very aware that this could be a good application for an abstract base class (especially since I expect to implement calculate_transfer_function and apply_inverse_transfer_function methods for planaranisoND_polND), but I've chosen to delay this for now in fear of premature abstraction.

I'm happy to chat in person anytime tomorrow. My immediate focus is on the interface for the "Retardance + Orientation" pipeline which I expect to require about 3/4 of my day tomorrow. After that PR, I am planning to start working on the recOrder CLI side and link everything up.

[talonchandler]

Thanks for the contributions and in-person chat @mattersoflight. I've made the following changes:

• renamed the models using the schema <object-type>_<object-thickness>_<data-shape>. We're starting with isotropic_thin_3d, phase_thick_3d, and inplane_anisotropic_thin_pol3d.
• fixed isotropic_thin_3d so that it handle absorption, and fixed the example to simulate a thin object.
• renamed from 2D/3D -> 2d/3d following PEP8 and one of Keith's comments in the dragonfly repo.

[talonchandler]

Merging into alg-dev...we can discuss more when I open a PR into main