pytorch_trainer and online dataset

[sjohn4]

Everything needed for generative tasks appart from the dataset, we can maybe also turn this into just pytorch trainer or just online_dataset.

[codecov]

Codecov Report

❌ Patch coverage is 90.90909% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.64%. Comparing base (c961257) to head (9c16c0a).

Files with missing lines	Patch %	Lines
.../trainer_server/internal/dataset/online_dataset.py	89.47%	6 Missing ⚠️
...trainer_server/internal/trainer/pytorch_trainer.py	92.50%	3 Missing ⚠️
...odyn/trainer_server/internal/dataset/data_utils.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #665      +/-   ##
==========================================
+ Coverage   85.58%   85.64%   +0.06%     
==========================================
  Files         258      258              
  Lines       11378    11449      +71     
==========================================
+ Hits         9738     9806      +68     
- Misses       1640     1643       +3     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

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

[sjohn4]

This basically changes the online dataset and pytorch trainer to be either to handle generative targets and non labeled outputs.
The online dataset basically now includes the possibility of receiving target samples that are also bytes like the normal samples and need either the same or different transformation functions applied. The pytorch trainer changes by having basically a branching path to handle generative training that expects longer outputs from the model. It also substitutes padding for -100 labels to be ignored in loss calculation.