This PR move the `ExpressionEvaluator` to `executor-utils` and
generalizes it such that it can evaluate an `AlgebraicExpression<T>` to
any type, not just `T`. This makes it possible to use the evaluator in
the backends. I used it in Plonky3 and Stwo, which leads to significant
code deletion.
The main feature of `ExpressionEvaluator` is that it handles
intermediate polynomials by caching during evaluation. This is cheaper
than using `Analyzed::identities_with_inlined_intermediate_polynomials`,
which might build exponentially large expressions. The Plonky3
implementation already did the same; The stwo implementation still used
`identities_with_inlined_intermediate_polynomials()` and now handles
intermediates properly.
With this PR, we compute the later-stage witnesses per machine instead
of globally. This has two advantages:
- We're able to handle machines of different sizes
- We can parallelize later-stage witness generation
This affects the two backend that can deal with multiple machines in the
first place: `Plonky3Backend` and `CompositeBackend`
