avm2: Implement ScopeChain caching

[Bale001]

Implements the ability for ScopeChain's to cache their results for better performance. Currently to avoid any problems with dynamic properties, caching is disabled for any ScopeChains that contain a with scope.

[Bale001]

BTW: If we ever implement an optimizer such as the one in #7920, we can actually do something even better than this since it would then be possible to resolve everything before the function is even called.

[adrian17]

BTW: If we ever implement an optimizer such as the one in #7920, we can actually do something even better than this since it would then be possible to resolve everything before the function is even called.

I'm not 100% sure that's true.
This is related what we talked about in the past - a lot of lookups we do (op_coerce/op_astype/op_istype or resolve_type) shouldn't use the scope chain at all - in fact, it's incorrect to do so and it may be possible to construct some evil code that breaks current Ruffle due to this. Types (and more?) should instead use some global (not sure if fully-global or abc-global or what) registry that's not recursive at all.
(this would also let us remove the resolve_class_private weirdness)

The main(?) difference from optimization POV is that most/all the lookups using this global registry can be fully resolved eagerly - like, you can resolve function arg types and store them forever. And this is not really an optimization, things totally rely on types not changing at runtime. However the "real" getlex/findproperty lookups might be able to optimize to getting a specific scope with getglobal/outerscope (not sure if even that it possible 100% of time), but you still need to getproperty/getslot the actual value at runtime, as the value can be modified at runtime.

EDIT: I made an "evil swf" where the Coerce opcode works on FP, but fails on Ruffle due to dynamic scope lookup finding a non-type. So yeah they definitely need to ne decoupled first.

[Dinnerbone]

Profiled the box2d demo before (red) and after (yellow) this change:

About 8% faster!