How cracy would it be to implement fn cost in Wasm?

A little bit crazy. I wouldn't do that just to get a hash of it.
One thing I've been thinking about is using a derive macro that just hashes the incoming source code.

How can we ensure that changes in is_accounting_operator (implemented in the metering middleware) would lead to a different hash? Can we get hash values from normal functions or would it need to be implemented as a macro?

Hm, that's a tough one. The macro can only access the code that's annotated with it.
But generally, this shouldn't be a problem for us, since we bump the serialization version with each wasmer upgrade anyways (if we want to get really crazy, we can even have a macro that reads our Cargo.toml and hashes the wasmer version).

So, basically there is https://docs.rs/code-hasher/0.1.0/code_hasher/ and https://docs.rs/include-cargo-toml/0.1.0/include_cargo_toml/, but we could also implement our own, since both of these are very small.

Sounds great. Happy for any implementation that works and that we can replace. So the idea here is to hash together 1. the fn cost source code and wasmer-middlewares version?

Yes, exactly.

Another note: We should make sure to backport this to all supported versions when it is implemented.