Is one supposed to grayscale and brightness contrast process the image before sending to textract?

Grayscaling and contrast-correcting images before sending to Textract should not be "required" and the service is trained and evaluated on a broad variety of OCR tasks.

However, as you've found, there are definitely cases where adding your own pre-processing if possible can help boost accuracy - sometimes with significant improvement. For example you might also explore skew detection and correction if that's relevant to your images, as I've seen customers benefit from it before.

It's worth mentioning that Amazon Textract is optimized more towards document extraction use-cases, while the Rekognition DetectText API may perform better in situations where the background is busier and there are fewer than 100 words to detect. I'm not sure how well Rekognition performs with handwriting (it's not explicitly mentioned in the doc from what I can see), but might be worth trying out to compare since your images seem to have relatively little text.

Grayscale and brightness contrast processing is not a pre-requisite for using Textract. There may be different use cases where pre-processing will produce better results but it is not required.