Some commits that should together fix the segfaults

[dschwoerer]

Includes #317 and #323

[dschwoerer]

Woohoo - it seems there are no segfaults anymore.

Time to fix the other bugs!

[dschwoerer]

Should we merge this, to make it easier fixing the remaining bugs?
#317 and #323 should probably be merged first ...

[ZedThree]

Change to h5netcdf should be fine. CI still has a lot of failures though -- and is it supposed to take an hour?!

[dschwoerer]

Yes, I think there are some regressions, that are only now noticeable that the segfaults are resolved.
They should be much easier to fix then the segfaults.

I do not know, I checked some python3.8 run from a few days ago, and that was 55min, 3.9 was 45 minutes.

Note that python 3.14 is hugely faster with 20 minutes. From a quick look it seems the errors are the same, so we might really benefit from removing the GIL lock, which would be nice.

[dschwoerer]

@ZedThree can you reproduce these issues? I tried on my fedora machine, but there everything works fine.
Maybe something is wrong with the h5netcdf wheel?