Fix flaky floating point comparison in PowerTransformY test (facebook#5044)

Summary:
Pull Request resolved: facebook#5044

`test_transform_optimization_config` is flaky because it uses strict equality to compare `OutcomeConstraint` objects after a `transform` -> `untransform` round-trip. The round-trip through `PowerTransformer` (which uses numerical optimization in `fit()`) can introduce small floating point errors in the constraint bound (e.g., `2.345` -> `2.3449999999999998`). Since `OutcomeConstraint.__eq__` ultimately compares string representations of bounds, even tiny differences cause assertion failures.

Replace the strict `assertEqual` with component-wise assertions that use `assertAlmostEqual` for the bound (tolerating FP error to 5 decimal places) while still checking all structural properties exactly.

See https://github.com/facebook/Ax/actions/runs/23172998708/job/67328952469 for an example failing job.

Reviewed By: Balandat

Differential Revision: D96855312

fbshipit-source-id: d75b56cb9a9bf3c2311c990a756d2a37f15f925b

Author: mpolson64

ax/adapter/transforms/tests/test_power_y_transform.py

@@ -251,7 +251,12 @@ def test_transform_optimization_config(self) -> None:
         cons = tf.untransform_outcome_constraints(
             outcome_constraints=oc_tf.outcome_constraints, fixed_features=None
         )
-        self.assertEqual(cons, oc.outcome_constraints)
+        self.assertEqual(len(cons), len(oc.outcome_constraints))
+        for c_actual, c_expected in zip(cons, oc.outcome_constraints):
+            self.assertEqual(c_actual.metric_names, c_expected.metric_names)
+            self.assertEqual(c_actual.op, c_expected.op)
+            self.assertEqual(c_actual.relative, c_expected.relative)
+            self.assertAlmostEqual(c_actual.bound, c_expected.bound, places=5)
         # Relative constraints aren't supported
         oc = OptimizationConfig(
             objective=objective_m2,