fix ema checkpoint loading

[JSabadin]

We already support loading hyperparameters such as lr_schedulers, epoch, global_step, optimizer_steps, and state_dict (including EMA weights if used) via the --resume flag in the train command. This PR:

1. Fixes a bug in the EMA implementation.
2. Ensures proper handling of the on_load_checkpoint hook, which is called before the on_fit_start hook.
3. Introduces a new parameter trainer.resume_training to resume training from checkpoints specified in the config file (model.weights).

[codecov]

Codecov Report

Attention: Patch coverage is 86.66667% with 2 lines in your changes missing coverage. Please review.

Project coverage is 95.76%. Comparing base (631b905) to head (fd09c8c).
Report is 35 commits behind head on main.

✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
luxonis_train/core/core.py	33.33%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #156      +/-   ##
==========================================
- Coverage   96.31%   95.76%   -0.56%     
==========================================
  Files         147      170      +23     
  Lines        6304     7552    +1248     
==========================================
+ Hits         6072     7232    +1160     
- Misses        232      320      +88     

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

[klemen1999]

LGTM