In order to implement infinite loop detection while executing MIR, both
the implementor of `Machine` (`Evaluator`) and its associated type
(`MemoryData`), must implement `Eq` and `Hash`. This PR adds the
required trait implementations.
It's possible that the `Hash` implementations need to be improved; only
the `env_vars` field of `Evaluator` and the `thread_local` field of
`MemoryData` are actually being hashed. Omitting fields from a `Hash`
implementation is not incorrect, but could lead to collisions if the
ignored fields are changing constantly. Perhaps I should instead derive
`Hash` on a few more fields related to MIR validation?
