Feat: add DipoleModel and PolarModel

[anyangml]

This PR is to provide model wrappers for DipoleFittingNet and PolarFittingNet, such that the saved model can be used directly in inference with DeepDiole and DeepPolar.

[codecov]

Codecov Report

Attention: 24 lines in your changes are missing coverage. Please review.

Comparison is base (4956864) 75.29% compared to head (5f1aec9) 75.31%.

Files	Patch %	Lines
deepmd/pt/model/model/dipole_model.py	63.88%	13 Missing ⚠️
deepmd/pt/model/model/polar_model.py	70.83%	7 Missing ⚠️
deepmd/pt/model/model/dp_zbl_model.py	42.85%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #3309      +/-   ##
==========================================
+ Coverage   75.29%   75.31%   +0.01%     
==========================================
  Files         398      400       +2     
  Lines       33694    33783      +89     
  Branches     1604     1604              
==========================================
+ Hits        25371    25442      +71     
- Misses       7462     7480      +18     
  Partials      861      861              

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

[anyangml]

@njzjz @wanghan-iapcm , not sure if my OutDef shape is wrong

I am expecting 1, 5, 3, 3 and 1, 5, 3, 9.
we may need to change this function to accommodate dipole shape.

Also, we have to add dipole_derv_c otherwise, we have length mismatch here

deepmd-kit/deepmd/pt/infer/deep_eval.py

Lines 245 to 251 in 4b994df

	)
	return dict(
	zip(
	[x.name for x in request_defs],
	out,
	)
	)

[anyangml]

this will return 0, since get_sel_type() returns an empty list in GeneralFitting.

deepmd-kit/deepmd/infer/deep_tensor.py

Line 107 in 4b994df

sel_natoms = self._get_sel_natoms(atom_types[0])

[njzjz]

@njzjz @wanghan-iapcm , not sure if my OutDef shape is wrong
I am expecting 1, 5, 3, 3 and 1, 5, 3, 9. we may need to change this function to accommodate dipole shape.

We have to use torch.permute to change the output shape, so it can be compatible with the current DPLR code. (obviously, we don't want to rewrite these codes or break the existing TF models)

For how OutDef defines, I will follow @wanghan-iapcm's suggestion.

[njzjz]

this will return 0, since get_sel_type() returns an empty list in GeneralFitting.

_get_sel_natoms needs to handle this particular situation.