Gradient operator - allow anisotropic scaling

[MargaretDuff]

For a BlockOperator we can scale each operator in the Block separately using __rmul__: https://github.com/TomographicImaging/CIL/blob/64e0b1b6b871de0a148adca3845104020609f438/Wrappers/Python/cil/optimisation/operators/BlockOperator.py#L359-L380

The GradientOperator acts like a BlockOperator, returning a block data container, but crucially isn't and thus does not allow this scaling, i.e. if you attempt to scale by a np.array you end up with an array of Scaled operators

This came from a discussion with Martin Sæbye Carøe about anisotropic gradient regularisation, and how the GradientOperator is much quicker than the finite difference operator but doesn't allow this anisotropic scaling.

[MargaretDuff]

Discussed in the dev meeting this morning. To get around the question of it should scale like an Operator or like a BlockOperator could be avoided if we added a "weight" or "scaling" argument (similar to weightedleastsquares or leastsquares)