Add probabilistic iterative methods

[gw265981]

What's the new feature?

As an example shows, there is potential in running the coreset algorithms that have a refine method iteratively. In particular, probabilistic KH can be run iteratively and with varying temperature (probability parameter).

We should decide whether this should be implemented as:

• A separate class that iterates an existing solver with a refine method (and some extra parameters),
• Part of an existing class, e.g., KernelHerding, as a parameter that allows iteration,
• Be left as an example for the users.

What value does this add?

This could produce novel coreset methods that are useful in some scenarios.

Is there an alternative you've considered?

No response

Additional context

Related to #953, #941, #244.

[gw265981]

Blocked by #947 for now as it introduces a version of this.