feat: pretraining dataset validation

[MichaelTj02]

Summary

Implemented a pretraining dataset validation step that blocks training initialization when the dataset is invalid and provides any detected dataset issues to the user.

Type of change

• Feature
• Fix
• Other (docs, refactor, chore, ci, etc.)

How was this tested?

• Launch Autolume
• Set the dataset path in the training module to a folder containing test data
• Start training with both a valid and an invalid dataset
• Verify that training proceeds normally for a valid dataset
• Verify that training is blocked for an invalid dataset and that dataset validation issues are displayed before the training initialization phase begins

2026-06-03.23-27-03.mp4