This PR changes the automation in `deriving_LawfulEq_tactic_step` to use `with_reducible` when asserting the shape of the goal using `change`, so that we do not accidentally unfold `x == x'` calls here. Fixes #10416.