Update Prithvi-v2-BurnScars Checkpoint to Support Fine-Tuning

This PR updates the Prithvi-v2-BurnScars checkpoint to include optimizer_states, which are required for continuing fine-tuning. The previous checkpoint from HuggingFace lacked this field because it was saved with weights_only=True.

• Regenerated compatible checkpoints using the same configuration
• Verified that fine-tuning now works on the custom data module

I removed the epoch, global_step, loops, callbacks, since these fields cause the training to resume from the best previous checkpoint (i.e. if you set max trainer.max_epochs=50, you might only get 10 epochs if the best previous checkpoint was achieved at epoch 40).