Production-ready training script with all robustness improvements

Robust fix for LoRA training - disable gradient checkpointing and fix gradient flow

Ensure train.py uses eval_strategy not evaluation_strategy

Add debug output to capture training errors

Restore training app with fixed train.py

Fix GPU device placement issue

Add diagnostic app to identify training issues

Add debug output to capture training errors

Fix Gradio Timer issue - use manual refresh instead

Add training app with Flan-T5 implementation and datasets

initial commit