remove StoreContextMut::with_{a,de}tached_instance[_async]

[dicej]

This is the second phase of the cleanup I started in an earlier commit with the goal of reducing unsafe code in the concurrent module and submodules, especially with regard to functions which take mutable references to a store and a ComponentInstance within that store.

My earlier attempt to make such functions safe was to temporarily "detach" the instance from the store to avoid aliasing hazards, but that got messy and confusing, and Alex and I decided an index-based approach made more sense. So now we pass Instance handles around instead of ComponentInstance references and pointers, using them to retrieve the ComponentInstance from the store only as necessary.

Practically speaking, this required moving a lot of inherent functions from ComponentInstance to Instance and updating their bodies accordingly. Functionally, nothing has changed.

[alexcrichton]

Very nice, thanks again for pushing through this!

Two minor comments:

• We probably want to keep an eye out for ergonomics here as some of the concoctions of .0 or .store_opaque_mut() or such can get pretty wordy. I'll try to go over this tomorrow and see if I have any ideas.
• Initially I was surprised about the use of Instance in most places as opposed to &Instance. That's often an antipattern in Rust b/c it can copy quite a few bytes, but here Instance is so small that it looks like it gets passed in registers on 64-bit platforms anyway so seems reasonable to me.

[dicej]

Yeah, I went back and forth about &Instance vs Instance. I'm fine with whatever.

Also agreed about the ergonomics of looking up ComponentInstance. One thing that occurred to me is to define a trait that StoreContextMut, &mut StoreOpaque, and &mut dyn VMStore all implement and make Instance::instance take a generic param bounded by that trait. i.e. Rust-style "overloading".