- Minimal Kronecker kernel. The inverse is highly inefficient - this can be improved!
- Added white noise kernel.
- Added diagonal computation abstraction.
- Added minimal computation abstraction for kernel combinations (i.e., for * and +).

Main issues:
- Though we can replace the computations in the gps.py, we might miss out with caching stuff e.g., the Cholesky decomposition in the dense computation case. Is there a sensible way around this?
- As a further point to the preceding point, we should not be computing the gram matrix in general (yes I known we need to in the dense setting).

- Think about computation savings for KL divergences - how can we resolve these?

Minor issues:
- Need to discuss the best way to add some computation methods - it might be tidier than just adding a load of static methods altogether.
- Need tests.

Need to add/update tests/ check code is correct.

Need to write complete tests for covariance operators, then can get on with revamping the rest of the codebase and tests for other modules.

Missing matmul, and need to improve the solve tests to ensure there are now issues with shapes.