Experiment with `PrunedStore` for Wasmi's executor

[Robbepop]

Currently Wasmi's bytecode executor is in parts generic over a host provided T which comes from the Store<T> where T represents the host data with which the host side can interact in host function calls.

Wasmi executor being generic in parts has big implications:

1. Codegen: The compiler might codegen multiple versions of big parts of the executor which is a huge component in Wasmi.
2. Tail Calls: Using tail call dispatch is not feasible as long as parts of the bytecode executor are still generic since it is hard to have a single source of execution handlers (and their function pointers).
3. FFI: Compiling Wasmi to a shared object and then using it is very complicated. Users have to work around that fact. This is especially true for Wasmi's C-API and future projects such as Wasmi's Python bindings.

The trick that might work is a PrunedStore that drops the T generic parameter by instead storing a core::any::TypeId in its place. And when converting back to a Store<T> on the host side (because the host knows its type) the TypeId can be checked against the T. If they are not equal we can simply error or panic and thus the system remains safe.

This way Wasmi's executor can use a PrunedStore instead of Store<T> internally and let the host side do the conversions. Though those conversions need to be hidden from the actual user of the Wasmi library.