Improve the check_generics pass

[Nadrieril]

This does a few things to improve the check_generics pass:

• We now track for each GenericArgs what item it applies to. This greatly simplifies the pass and will be a tremendous help for Hide associated types equality constraints #127;
• We pretty-print the llbc before running the pass, which makes it possible to inspect the llbc output even when it faile;
• We now track which items have been visited so that errors have more context to them.

[Nadrieril]

Other features I'd like to eventually add to this pass:

• Check that each generic variable reference points to something;
• Check the number and type of function arguments in function calls (will likely require fixing closures first);
• Check that referenced associated types actually exist (I'm waiting for Support for Generic Associated Types #477);
• Check the number and type of aggregate fields (also waiting for closure fixes);
• Check the generics and arguments of built-in types and functions.

[sonmarcho]

Checking that I understand what you are doing: the GenericSource tracks the item owning the generic args, correct?
For instance, if I have Array::index<T, N> then the generic source of <T, N> is Builtin, while it is Item if I have Vec<T>, right?
Whenever I need this kind of information in a micro-pass, I implement a visitor which carries an (optional) environment around, by overriding the proper methods to update the environments whenever I, say, visit a Call, Adt construct, etc. I'm wondering: why didn't you make that choice (it may be more straightforward/natural to do in OCaml of course)?

[Nadrieril]

For instance, if I have Array::index<T, N> then the generic source of <T, N> is Builtin, while it is Item if I have Vec, right?

Correct

Whenever I need this kind of information in a micro-pass, I implement a visitor which carries an (optional) environment around, by overriding the proper methods to update the environments whenever I, say, visit a Call, Adt construct, etc. I'm wondering: why didn't you make that choice (it may be more straightforward/natural to do in OCaml of course)?

This is pretty much what I did before, you can see the diff in the PR. The problem is that this requires manually listing every spot where a GenericArgs might be, which is error-prone and annoying to maintain. In fact my previous attempt at simplifying this was by using FunDeclRef/TraitDeclRef/etc consistently, but this wasn't enough for e.g. function calls without other undesirable changes.

This would have been fine if it was just for this pass, but #127 requires something similar and even more subtle, hence why I came up with this imo very pleasing solution.

[sonmarcho]

That makes sense, thanks for the explanations!