Confusing UserWarning when using `CouplingFlow`: Gradients do not exist for variables

[vpratz]

Using CouplingFlow currently raises a warning from Keras, see for example this test log. As far as I can tell, it does not impact functionality, because we calculate the loss in a different fashion. Is this correct? If so, I think finding a way to suppress this warning would be great to avoid confusion.

tests/test_two_moons/test_two_moons.py::test_fit[inference_network='coupling_flow']
  /usr/share/miniconda/envs/test/lib/python3.11/site-packages/keras/src/optimizers/base_optimizer.py:735:
UserWarning: Gradients do not exist for variables ['coupling_flow_120/dual_coupling_1152/single_coupling_2304/mlp_3064/configurable_hidden_block_6752/projector', 'coupling_flow_120/dual_coupling_1152/single_coupling_2305/mlp_3065/configurable_hidden_block_6754/projector', 'coupling_flow_120/dual_coupling_1153/single_coupling_2306/mlp_3066/configurable_hidden_block_6756/projector', 'coupling_flow_120/dual_coupling_1153/single_coupling_2307/mlp_3067/configurable_hidden_block_6758/projector']
when minimizing the loss. If using `model.compile()`, did you forget to provide a `loss` argument?