add @version to serialization data

[njzjz]

At this time, all current versions and the minimal and maximum versions are set to 1. If any changes, whether breaking or not, are made to the serialization data, the version should be bumped.

[codecov]

Codecov Report

Attention: Patch coverage is 98.50746% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 75.69%. Comparing base (004ebd6) to head (47bfd4a).

Files	Patch %	Lines
deepmd/utils/version.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #3349      +/-   ##
==========================================
+ Coverage   75.64%   75.69%   +0.04%     
==========================================
  Files         411      412       +1     
  Lines       34569    34635      +66     
  Branches     1599     1603       +4     
==========================================
+ Hits        26151    26216      +65     
- Misses       7558     7559       +1     
  Partials      860      860              

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

[wanghan-iapcm]

What kind of convention shall we adapt to numbering the model version?
shall we support separate version control for fine-grained modules? imagine the case that the descriptor of a model may be used to initialized the descriptor of another model.

[njzjz]

What kind of convention shall we adapt to numbering the model version?

The version should be bumped when compatibility is broken with either forward or backward.

shall we support separate version control for fine-grained modules? imagine the case that the descriptor of a model may be used to initialized the descriptor of another model.

I think we should support separate versions for each serialize and deserialize method, which is the most convenient for developers.