Update NaN filtering in InterpolatingMap

[JelleAalbers]

What does the code in this PR do / what does it improve?

Fixes InterpolatingMap so we do not crash with scipy 1.11.

Can you briefly describe how it works?

Scipy 1.11's kdtree no longer accepts nonfinite values (scipy/scipy#18502). However, we sometimes use NaN to represent missing/undefined values, e.g. for drift time if an event has no S1.

Currently we pass all points to InterpolatingMap, then force the result to NaN for points that have nonfinite distance to the neighbours the kdtree finds. That's fine in our current scipy 1.10 environment, but once we update (XENONnT/ax_env#353) there will be problems.

This PR instead checks for NaN before passing the points to InterpolatingMap to avoid crashing.

Can you give a minimal working example (or illustrate with a figure)?

See the new unit test.

I tried processing a small run 026195 locally with scipy 1.11.3, and it crashed here

straxen/straxen/plugins/events/event_positions.py

Line 164 in 933cc54

delta_r = self.map(corr_pos)

where events with NaN drift time are being passed to a map. Now it no longer crashes.

[dachengx]

Thanks @JelleAalbers . Just for my education. How did you find this potential future crash?

[JelleAalbers]

Thanks Dacheng -- I was processing a tiny run on my laptop, which meant all the package versions were different from the ones we usually use.

[dachengx]

Thanks @JelleAalbers .