pt: Add support for dipole and polar training

[iProzd]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 91.20000% with 11 lines in your changes are missing coverage. Please review.

Project coverage is 76.66%. Comparing base (59d3b12) to head (7c1c978).

Files	Patch %	Lines
deepmd/pt/infer/deep_eval.py	28.57%	5 Missing ⚠️
deepmd/dpmodel/infer/deep_eval.py	33.33%	4 Missing ⚠️
deepmd/pt/loss/tensor.py	96.55%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #3380      +/-   ##
==========================================
+ Coverage   76.59%   76.66%   +0.06%     
==========================================
  Files         420      421       +1     
  Lines       35754    35838      +84     
  Branches     1620     1620              
==========================================
+ Hits        27387    27476      +89     
+ Misses       7499     7494       -5     
  Partials      868      868              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[njzjz]

I got inf when doing dp test:

/home/jz748/codes/deepmd-kit/deepmd/entrypoints/test.py:996: RuntimeWarning: divide by zero encountered in scalar divide
  rmse_fs = rmse_f / np.sqrt(sel_natoms)
/home/jz748/codes/deepmd-kit/deepmd/entrypoints/test.py:997: RuntimeWarning: divide by zero encountered in scalar divide
  rmse_fa = rmse_f / sel_natoms
DEEPMD INFO    # number of test data : 80
DEEPMD INFO    Dipole  RMSE       : 1.437255e+00
DEEPMD INFO    Dipole  RMSE/sqrtN : inf
DEEPMD INFO    Dipole  RMSE/N     : inf

The number of atoms seems to be based on sel_natoms?