SimpleGenerativeModel: parametrize scheduler

[ArneBinder]

This PR adds the optional parameters scheduler_name and scheduler_kwargs to SimpleGenerativeModel which allows to select a training scheduler different then then the linear one. Internally, transformers.get_scheduler is now used to set up the scheduler (instead of using transformers.get_linear_schedule_with_warmup). Note that the default behavior is kept if scheduler_name is not specified (i.e. if warmup_proportion > 0.0 use linear with warmup otherwise don't use any scheduler).

See transformers.optimization.TYPE_TO_SCHEDULER_FUNCTION for available options for scheduler_name.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 80.00000% with 2 lines in your changes missing coverage. Please review.

Project coverage is 95.46%. Comparing base (6abf673) to head (4a8c0c3).

Files with missing lines	Patch %	Lines
src/pie_modules/models/simple_generative.py	80.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #148      +/-   ##
==========================================
- Coverage   95.49%   95.46%   -0.04%     
==========================================
  Files          61       61              
  Lines        5173     5179       +6     
==========================================
+ Hits         4940     4944       +4     
- Misses        233      235       +2     

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