Each time we allocate an object from C++ code, we go through a lot of effort to figure out the AllocKind, prototype, Shape etc to use for the new object. The NewObjectCache mitigates this a little but it's not very robust and has its own issues.

Furthermore, in NewObject we have special code for JSFunction and TypedObject. These really want their own optimized allocation path instead of slowing down the other cases.

We should experiment with a faster allocation path for JSFunction, ArrayObject, etc based on the shape we can cache on the global.