Allow evaluation module to be specified when building an ODEFunction

[dpad]

This adds the ability to manually specify which module's cache to use when an RGF is built as part of an ODEFunction. Basically, together with eval_expression you can now specify eval_module=@__MODULE__ (the default, which simply uses ModelingToolkit's RGF cache).

I've added an example test case, which consists of building an ODEFunction as part of a precompiled user module.
In this case, "f_bad" throws a KeyError because it does not use eval_module, so it's tagged in the ModelingToolkit RGF cache which is not kept after precompilation of the user module is done.
On the other hand, "f_good" works because it supplies the user module to eval_module, allowing ModelingToolkit to build the RGF within the user module's cache.

NOTE: there's a few other uses of @RuntimeGeneratedFunctions that probably could do with a similar fix to this, or some better solution. I am not sure how best to tackle this other than the current way, so I would appreciate any comments, but I'll leave that out of scope for this PR as it at least fixes #678 for me.

[ChrisRackauckas]

I think this should work. I'll just want @c42f's opinion on the safety.

[c42f]

Perhaps there's two separate things desired here?

1. Ability to specify which module's precompile cache the RGF lives in
2. Ability to specify which module is used to look up global symbols used in that RGF (including generic functions)

In principle these might be separate modules, I suppose, so SciML/RuntimeGeneratedFunctions.jl#20 might only be half the story?

[ChrisRackauckas]

Oh I just read this. Yes, good point.

[dpad]

Updated this to work with SciML/RuntimeGeneratedFunctions.jl#20.
NOTE: I changed the compat requirements for RuntimeGeneratedFunctions.jl to 0.4.3 which I assumed will be the next tagged release including the change of SciML/RuntimeGeneratedFunctions.jl#20.

[ChrisRackauckas]

Can this get updated for SciML/RuntimeGeneratedFunctions.jl#22 and handle all of the system types?