Refactoring sample from dicts to named arguments + **kwargs

[gvegayon]

This PR includes the following:

• Updates the metaclasses RandomVariable and Model to use **kwargs for sample().
• Updates all instances of these under the submodules.
• Within the model module, it ensures that arguments are via keywords instead of dictionaries.
• Updates the tests to reflect the change.
• Updates the demos under docs to reflect the change.

For a new discussion: Now that I've done this refactoring, the need to standardize the submodules became apparent.

• When specifying the observation process for any of the models, it makes sense to have the RandomVariable.sample() method of observed class feature mean and obs. This way the HospitalizationModel would always expect the observation process to receive mean for the latent hospitalizations and obs for the observed hospitalizations.
• Other arguments could still be passed as now **kwargs are unpacked whenever calling a RandomVariable.sample() method.
• Like the observed, process could also have defaults for RandomVariable.sample(), e.g., n_timesteps.

[AFg6K7h4fhy2]

Want me to review this? So 47 is not bottlenecked?

[dylanhmorris]

Some tweaks.

RE: Observation classes: as discussed in person, we should standardize the name of the predicted value and observed value keyword arguments to an Observation's sample() method. I think obs for observed (to be consistent with numpyro.sample(). For the prediction, how about pred, which is nicely consistent with obs?

GIven this, I agree that more templating makes sense. Time for an abstract Observation(RandomVariable) metaclass?

[gvegayon]

@AFg6K7h4fhy2 and @dylanhmorris, all done with your requests:

• Improved doc on **kwargs: I still left a couple where it says Ignored, but that's because it is literally that.
• Renamed mean to predicted for the observation processes.
• Replaced a default value RandomVariable to none to avoid mutable defaults (but left a few strings, as strings are inmutable).

[AFg6K7h4fhy2]

Approve conditional on some minor comments being addressed.

No big changes since last check. Most unaddressed issues will be resolved with existing issues+PRs.