Minor fix to Donut SROIE training

[alienware]

WHAT

• Import to_pil_image from ImageFeatureExtractionMixin instead of processor.feature_extractor
• Fixes sample and test_sample references

WHY

Fixes AttributeError: 'DonutImageProcessor' object has no attribute 'to_pil_image'

[alienware]

First of all, lots of thanks to you for providing these notebooks and for your writeups. These are a blessings in picking up document AI. 🙏

The fix I'm proposing may not be the ideal one, but what I tried after I couldn't get the to_pil_image statement to work. I'm assuming I am facing this error due to a different transformers version.