NeuralOperator Implementation

[samuelburbulla]

The current NeuralOperator implementation (and related stuff) is only prototypical and needs revision. Many things can be implemented more generally.

• [x] NeuralOperator implementation should be generalized to handle arbitrary Operator layers.
• [x] NeuralNetworkKernel is currently a convolution k(x - y), but should be more general.
• We could also implement a point sampling strategies to enable different function evaluations in the layers of a neural operator.

[samuelburbulla]

Most of the work is related to #23