According to our benchmarks those string type wrappers and their custom Ord implementation improve Linker performance by roughly 50%. This is probably because the Rust compiler is able to inline the `memcmp` routines whereas with `Arc<str>` and `str` it uses `memcmp` calls which are inefficient for small strings.

The new implementation optimizes the Ord impl of ImportKey since ImportKeys are compared more often than their fields are queried during Linker setup.

This allows us to create new HostFuncTrampolineEntities without an Engine and thus allow for a HostApi abstraction later to further optimize Linker setup.
Also conveniently this allow us to get rid of tons of custom Debug implementations to resolve the no longer existing DedupFuncTypes.