Consistent activation functions between backends

[njzjz]

1. add relu, gelu, gelu_tf, relu6, softplus, sigmoid, and linear to dpmodel;
2. add gelu_tf, relu6, soft6, softplus, and sigmoid to pt;
3. change gelu in pt from non-approximate to approximate. If one still wants to use the non-approximate version, we may consider to add a new key;
4. add linear to tf;
5. none in tf now returns lambda x: x instead of None to be type consistent;
6. support uppercase in all backends;
7. add consistent tests.

[codecov]

Codecov Report

Attention: Patch coverage is 98.00000% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 77.08%. Comparing base (09bd522) to head (ca004b4).

Files	Patch %	Lines
deepmd/dpmodel/utils/network.py	96.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #3431      +/-   ##
==========================================
+ Coverage   76.98%   77.08%   +0.09%     
==========================================
  Files         427      394      -33     
  Lines       36753    31991    -4762     
  Branches     1621      909     -712     
==========================================
- Hits        28295    24659    -3636     
+ Misses       7590     6804     -786     
+ Partials      868      528     -340     

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