Improvements to FrontEnd models

[billbrod]

This PR includes some small changes to the FrontEnd models and related code:

• The std arg of gaussian1d can now be an int, and includes more validation
• Removes width_ratio_limit from CenterSurround, along with clamping behavior. That wasn't working and is only really needed when training the model, which isn't the focus of plenoptic. I think that sort of clamping behavior is better implemented in the training loop
• All the FrontEnd models have the following changes:
  • Remove width_ratio_limit, as described above
  • kernel_size can be an int or tuple of ints (also in Linear, Gaussian, CenterSurround models)
  • adds bit to documentation describing model name in Berardino's eigendistortion paper
  • Adds pretrained and cache_filt to all models, with parameters from Berardino's thesis (requires kernel_size=31 or kernel_size=(31,31)
• corrects same_padding type annotations to match behavior
• adds tests of above behavior

Closes #265, #264

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files with missing lines	Coverage Δ
...enoptic/simulate/canonical_computations/filters.py	100.00% <100.00%> (ø)
src/plenoptic/simulate/models/frontend.py	95.55% <100.00%> (+1.16%)	⬆️
src/plenoptic/simulate/models/naive.py	99.03% <100.00%> (+0.87%)	⬆️
src/plenoptic/tools/conv.py	100.00% <ø> (ø)

... and 1 file with indirect coverage changes