Compute stat_sf() by panel instead of group

[teunbrand]

This PR aims to fix #4340.

The most minimal solution I could come up with is to change the compute_group method to the compute_panel method. As far as I could tell, there is no summarising happening over groups, so I think this is safe. I'm not too familiar with {sf} though, so I'd appreciate a second pair of eyes here.

[thomasp85]

LGTM - have you done testing to assure that no use cases of sf plotting are affected by this switch to panel based stat-calculations? I can't envision what should fail, but just want to make sure

[teunbrand]

I'm not a very regular user of the sf functionality, so I'm having a hard time anticipating unwanted side-effects and what I should test for. Of course, the wanted side-effect is that now drawing order is preserved from input, rather than by group, so some things are different on purpose. I imagine if people are drawing maps with polygons that touch, the draw order would influence which line colour would be visible at the shared border.

[teunbrand]

I'm going to make a judgement call here and suppose this wouldn't affect any legitimate use-cases. I hope that in the bad-case scenario that it does, people will find out on the dev branch before a release.