Load Pretrained Models in tests using ZnTrack

[M-R-Schaefer]

          This could be replaced by `data = zntrack.from_rev(,,,)` followed by `ase.io.write(data.atoms, "file.xyz")`.

It probably doesn't matter for this one, but we could also use pre-trained models for testing this way.

One benefit would be, that this would allow running tests offline by altering the remote through a global variable:

e.g. export TEST_DATA_REMOTE=/path

I want to eventually do that in IPSuite and wanted to leave this as a suggestion her.
This would certainly be not fixed in this PR but in a further one.

Originally posted by @PythonFZ in #200 (comment)