Use builtin Astropy model unit and bounding box support

[lpsinger]

Define the input_units, return_units, input_units_equivalencies, and bounding_box properties for all models. Use Astropy models' built-in unit conversion support.

All models now require inputs with valid units (wavelength, wavenumber, or frequency). Dimensionless inputs are no longer automatically converted to wavenumber.

Fixes #237.

[lpsinger]

Note that there is one unit test that is failing due to astropy/astropy#16906.

[karllark]

Can you say how the valid x range in checked with these changes? Just can't see how this is done and figured I'd just ask.

[karllark]

And the unit test failing will need to be addressed.

[lpsinger]

Can you say how the valid x range in checked with these changes? Just can't see how this is done and figured I'd just ask.

I overrode the ``astropy.modeling.Model.prepare_inputsmethod inBaseExtModel` because it turns out that Astropy itself doesn't actually do bounds checking based on the bounding box except in rare cases like Tabular1D.

[lpsinger]

And the unit test failing will need to be addressed.

Yup. Waiting to hear back from upstream about astropy/astropy#16906.

[karllark]

Can you say how the valid x range in checked with these changes? Just can't see how this is done and figured I'd just ask.

I overrode the ``astropy.modeling.Model.prepare_inputsmethod inBaseExtModel` because it turns out that Astropy itself doesn't actually do bounds checking based on the bounding box except in rare cases like Tabular1D.

That is clever! Neat.

[lpsinger]

I misunderstood what the bounding box is for. It's the "limits over which the model is significant", i.e., far from zero. However, the x_range property of the dust extinction models is really describing the limits over which the model is known or defined. These are not the same thing.

I'm closing this PR. Instead, here's one that just moves the input preparation to the base class and leverages the builtin model input units handling: #240.

See https://docs.astropy.org/en/stable/modeling/models.html#efficient-model-rendering-with-bounding-boxes:

All Model subclasses have a bounding_box attribute that can be used to set the limits over which the model is significant. This greatly improves the efficiency of evaluation when the input range is much larger than the characteristic width of the model itself.