change to almostEqual in UT to avoid difference at FP precision

[amcadmus]

No description provided.

[amcadmus]

cherry-picked from master

[codecov-commenter]

Codecov Report

Merging #1048 (02789b8) into devel (37af1d1) will decrease coverage by 18.62%.
The diff coverage is n/a.

❗ Current head 02789b8 differs from pull request most recent head 46f7502. Consider uploading reports for the commit 46f7502 to get more accurate results

@@             Coverage Diff             @@
##            devel    #1048       +/-   ##
===========================================
- Coverage   82.90%   64.28%   -18.63%     
===========================================
  Files         120        5      -115     
  Lines       10139       14    -10125     
===========================================
- Hits         8406        9     -8397     
+ Misses       1733        5     -1728     

Impacted Files	Coverage Δ
deepmd/descriptor/se_a_ef.py
source/lib/tests/test_soft_min_switch_virial.cc
deepmd/utils/__init__.py
source/lib/tests/test_main.cc
deepmd/fit/wfc.py
deepmd/model/model_stat.py
deepmd/cluster/local.py
source/lib/tests/test_pair_tab.cc
deepmd/utils/convert.py
source/api_cc/tests/test_deeppot_model_devi.cc
... and 105 more

---

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 37af1d1...46f7502. Read the comment docs.

[njzjz]

Do we want to merge master back to devel? If so, I think we don't need to cherry-pick.