[Fix] Assure Learning Rate Scheduler does not increase the learning rate

[ivallesp]

When the base learning rate is smaller than the minimum learning rate, the following operation is applied:

max(min_lr, factor*base_lr)

This effectively increases the learning rate once the patience epochs without improvement are reached.

Description of changes:

I included an assert statement to break in case min_lr > base_lr

[szha]

Job PR-359/1 is complete.
Docs are uploaded to http://gluon-ts-staging.s3-accelerate.dualstack.amazonaws.com/PR-359/1/index.html

[szha]

Job PR-359/2 is complete.
Docs are uploaded to http://gluon-ts-staging.s3-accelerate.dualstack.amazonaws.com/PR-359/2/index.html

[codecov-io]

Codecov Report

Merging #359 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff            @@
##           master    #359      +/-   ##
=========================================
+ Coverage   80.09%   80.1%   +<.01%     
=========================================
  Files         146     146              
  Lines        8356    8357       +1     
=========================================
+ Hits         6693    6694       +1     
  Misses       1663    1663

Impacted Files	Coverage Δ
src/gluonts/trainer/learning_rate_scheduler.py	100% <100%> (ø)	⬆️

[lostella]

@ivallesp good catch, thanks for the PR!