Better testing of serialization between Noir and TS

[benesjan]

We don't sufficiently test that the serialization in Noir matches that of TS.

I propose the following to address this:

1. For each struct which has both Noir and TS representation check that serialization of non-empty instance results in the same value.
2. Move the serialization functionality which is used when creating the witness to the classes to minimize probability of forgetting to do a change when changing something (e.g. functions like toACVMFunctionData or toACVMCallContext should be in the classes).

### Tasks
- [ ] For each struct which has both Noir and TS representation check that serialization of non-empty instance results in the same value.
- [ ] https://github.com/AztecProtocol/aztec-packages/issues/4332

[benesjan]

Partially tackled and the rest replaced with #4412