For adaptive localization output calculated correlations

[oyvindeide]

Make an optional datastructure that can output the calculated correlations in adaptive localization.

[oyvindeide]

@dafeda suggested we look to for example: https://github.com/bayesian-optimization/BayesianOptimization/blob/master/examples/advanced-tour.ipynb for inspiration on how to do this. They have a system of callbacks. We discussed it a bit, and this should be optional, as we probably dont want to do this for fields for examples.

[tommyod]

Some thoughts on this:

• A callback seems like a good option! I imagine either assimilate(..., corr_callback=None) or a new method subscribe where we take inspiration from BayesianOptimization.
• The callback could have signature callback(corr_XY:ndarray) or callback(corr_XY:ndarray, threshold:float). The caller then stores this information to disk within the callback function, or does something else with it.

This seems simple to implement. But beware that corr_XY can be huge, and that each entry is subject to a lot of noise since the number of realizations is so small.

[dafeda]

Want to fix #7941 first since I need a working jupyter notebook for easy debugging.

[dafeda]

Done