Replace random_split with Subset and add image path validation

[agentksimha]

Changes:

• Replaced random_split with Subset using explicit index slicing for deterministic and reproducible dataset splits
• eval_dataset now correctly wraps dataset_val ensuring validation-specific transforms are applied separately from training transforms
• Added os.path.isfile() guard in generate_text_from_image_segment to handle invalid or missing image paths gracefully instead of crashing with an unhandled exception

[abhiram123467]

“I’ll work on this – will add a detailed README for the OCR‑1 subfolder.”