ruff

c46acee0 - [ty] Refactor constraint traversal and cache satisfiability checks (#29492)

[ty] Refactor constraint traversal and cache satisfiability checks (#29492) This PR extracts some refactoring changes out of #29027 so they can be reviewed separately. This should be entirely behavioral neutral. The changes: - There is now a clear separation between "atomic" constraints and non-atomic constraints. There aren't actually any non-atomic constraints yet, but #29027 will add existential nodes as a non-atomic constraint. The defining characteristic is that an atomic constraint is one that is an actual assignment to a typevar; it is what `PathAssignments` builds up as it walks a BDD, and is what can live as an antecedent or consequent in a `SequentMap`. - We've extracted the logic for persisting a `ConstraintSet` into an `OwnedConstraintSet`, moving it into a standalone file and structuring it better via a helper type and some finer grained methods. - Removed some code sharing in `SolutionWalker` that was actually leading to more complex code. `visit_node` and `node_is_satisfiable_on_path` no longer use a shared helper, since the two uses had many differences, which required overridable hook points. The shared part was much smaller, so it's simpler to just copy/paste that part. - That makes it easier to add caching of `node_is_satisfiable_on_path`, similar to the caching we already had for `visit_node`.