Allow to scale LR in different ways.

[shishaochen]

In some cases, convergence speed of scale learning rate linearly can be worse than not scaling.
Thus, an argument is exposed for users.

[codecov-commenter]

Codecov Report

Merging #1167 (24b354c) into devel (ba087c4) will decrease coverage by 11.80%.
The diff coverage is n/a.

❗ Current head 24b354c differs from pull request most recent head 7e244e4. Consider uploading reports for the commit 7e244e4 to get more accurate results

@@             Coverage Diff             @@
##            devel    #1167       +/-   ##
===========================================
- Coverage   76.08%   64.28%   -11.81%     
===========================================
  Files          91        5       -86     
  Lines        7226       14     -7212     
===========================================
- Hits         5498        9     -5489     
+ Misses       1728        5     -1723     

Impacted Files	Coverage Δ
deepmd/utils/errors.py
deepmd/train/run_options.py
source/op/_prod_force_se_a_grad.py
deepmd/loss/__init__.py
deepmd/utils/network.py
deepmd/utils/random.py
deepmd/entrypoints/transfer.py
deepmd/utils/tabulate.py
deepmd/entrypoints/config.py
deepmd/op/__init__.py
... and 74 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 ba087c4...7e244e4. Read the comment docs.