Set up precompilation workloads

[MasonProtter]

It'd be good to add some precompilation workloads using https://github.com/JuliaLang/PrecompileTools.jl, so that we can reduce the amount of compilation time users experience the first time they solve various models.

This is more or less impossible to do effectively for the MTK backend, but is in principle possible / profitable with GraphDynamics. In practice, I've so far seen limited gains from doing it due to invalidations. E.g. I saw the compilation times for Anthony's PING tutorial solve call get reduced from ~10 seconds to ~6 seconds, which makes me a little unsure if it's worth making the Neuroblox take ~10 seconds longer to precompile when you first install it (or when you load it after modifying it).

Keeping this open as a reminder for myself to look into this more and see what I can do to get rid of these invalidations.