[ENH] Implement the Proximity Forest 2 classifier using aeon distances

[TonyBagnall]

Describe the feature or idea you want to propose

a new distance based classifier, Proximity Forest 2, has recently been proposed. This is probably the current best distance based algorithm for time series classification (we have not used it yet). IT would be fantastic to have an aeon implementation.

https://arxiv.org/abs/2304.05800

Describe your proposed solution

It should probably be done in conjunction with #159 (or instead of #159?) using aeon distance functions.

Describe alternatives you've considered, if relevant

No response

Additional context

No response

[hadifawaz1999]

I think it should be instead of #159 , unless someone already started to work on it, if not then yeah we can skip it and go directly to PF2

[GuiArcencio]

I can work on this since I've implemented PF before, I could have a working draft around next week or so. One possible issue for PF2 is that it requires different cost functions for DTW calculations (||a - b||^0.5, 1, or 2), which I think are fixed in our implementations.