Deprecate optimisers accessed through compat?

[njtierney]

They are currently supported but it seems that we might want to remove support as that's the direction it is going

https://www.tensorflow.org/api_docs/python/tf/compat/v1/train/

• adagrad_da
• proximal_gradient_descent
• proximal_adagrad