Forward shuffle to trainer server and measure GPU

[XianzheMa]

1. Add a unit test to how we pack the StartTrainingRequest, thereby catching the unspecified shuffle.
2. Enable accurate GPU operations measurement.

[codecov]

Codecov Report

Attention: Patch coverage is 86.66667% with 8 lines in your changes missing coverage. Please review.

Project coverage is 82.48%. Comparing base (e6541d3) to head (e1e3eb5).

❗ Current head e1e3eb5 differs from pull request most recent head 5588511

Please upload reports for the commit 5588511 to get more accurate results.

Files	Patch %	Lines
...trainer_server/internal/trainer/pytorch_trainer.py	77.41%	7 Missing ⚠️
modyn/common/grpc/grpc_helpers.py	87.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #526      +/-   ##
==========================================
+ Coverage   82.43%   82.48%   +0.04%     
==========================================
  Files         214      215       +1     
  Lines       10038    10049      +11     
==========================================
+ Hits         8275     8289      +14     
+ Misses       1763     1760       -3     

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

[github-actions]

^{( % to main)
( % to main)}

[MaxiBoether]

Thanks Xianzhe! I had a nitpicky naming comment, and I would like to always have the stopwatch running, even if we don't properly synchronize cuda.

[MaxiBoether]

Looks good!