Fix bugs in training pipeline

[gitttt-1234]

This PR fixes the following bugs:

• torch.load() in lightning_modules.py failed when using "gpu" as the device for map location, it should use "cuda" / "mps" / "cpu". Related Issue: Loading customized pretrained_backbone_weights throws PyTorch error #321

• Fixes a bug in computing the length of the dataloader. (If there's only one sample, the code failed to return 1)

• Resolved a compatibility issue where MediaVideo backends did not work with the threading model used for caching. (Removing the thread-based approach for now)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.46%. Comparing base (ff91433) to head (80a492d).
⚠️ Report is 29 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #322      +/-   ##
==========================================
- Coverage   95.28%   93.46%   -1.83%     
==========================================
  Files          49       49              
  Lines        6765     6946     +181     
==========================================
+ Hits         6446     6492      +46     
- Misses        319      454     +135     

☔ View full report in Codecov by Sentry.
📢 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.