Add enumerative inference to the inference library

[ztangent]

This PR introduces the enumerative_inference function, which can be used to perform:

• Exact inference over models with a finite number of discrete random choices.
• Grid approximation of continuous target densities (or mixed continuous-discrete targets)
• Stratified sampling by enumeration over discrete choices and sampling of continuous choices

The motivations for this PR are:

1. Making it easier to debug inference algorithms by having a (slower) gold-standard inference algorithm to compare against
2. Providing a simpler entry point for beginner users of Gen (e.g. students used to webPPL and trying to write probmods-style models). Such users often default to importance_sampling with lots of samples, even though there's no guarantee that gives you a good approximation of the posterior.

To use enumerative_inference, users provide an iterator over choice maps and their associated log-volumes, i.e. the log of the volume of sample space that each choice map is associated with (the volume is 1 if all choices are discrete). enumerative_inference then returns a trace and normalized log. probability for each (choices, log_vol) pair in the iterator, along with an estimate of the log marginal likelihood, essentially performing Riemann integration over the space of traces.

A grid of (choices, log_vol) pairs can be constructed using choice_vol_iter, by specifying addresses and a grid of values / intervals to enumerate over for each address.

Documentation has been added (see https://www.gen.dev/docs/previews/PR545/ref/inference/enumerative/), but I still might want to add a simple tutorial showing how enumerative inference can be used to debug a model in cases where importance sampling with the default proposal might be a give a poor approximation to the posterior (e.g. in cases where the observations are highly unlikely under the prior).

[ztangent]

Have decided to merge this for now and add the tutorial separately when I have more time! There is a usage example in the test cases (test/inference/enumerative.jl) that we can point people to in the meantime.