fix time behavior in tests and filter warnings for pandas 1.4

[CagtayFabry]

Changes

• Fix failing tests due to new pandas version ->seems like pandas added own __add__ functions so that we couldn't use + with a pandas Type on the left-hand side and our Time class on the right-hand side (thanks @vhirtham )
• filter warnings caused by xarray introduced with pandas 1.4

Checks

• updated tests

[CagtayFabry]

This should get rid of a few thousand warnings in tests hopefully 🚀😄
(and clean up your notebooks @marscher 😉) sorry it won't since this is only tests... 🙄 I suggest going back to pandas 1.3 for now

[codecov]

Codecov Report

Merging #689 (13e6bff) into master (d4c6039) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #689   +/-   ##
=======================================
  Coverage   95.89%   95.89%           
=======================================
  Files          92       92           
  Lines        6310     6310           
=======================================
  Hits         6051     6051           
  Misses        259      259           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d4c6039...13e6bff. Read the comment docs.

[github-actions]

Unit Test Results

       1 files  ±0         1 suites  ±0   2m 38s ⏱️ +19s
2 101 tests ±0  2 101 ✔️ +22  0 💤 ±0  0 ❌  - 22 

Results for commit 13e6bff. ± Comparison against base commit d4c6039.

♻️ This comment has been updated with latest results.

[marscher]

Eventually it makes sense to pin to a minor of pandas, which has been tested and only upgrade it, if everything works as expected.

[CagtayFabry]

something like pandas>=1.0,<=1.3 ?

I think that would be ok if we have a good workflow release patch versions if new pandas releases work as expected (while still giving some safety)

[marscher]

Yes.

PyPI prohibits uploading patched sources, so this always renders a new micro release just for updating the dependencies. Conda does circumvent this by increasing build numbers. I guess there is no way to avoid this.

Eventually it is too much overhead.