Do the uploading ourselves for textract so we can avoid repeated uploads

[eric-anderson]

The textract uploading support uses the default textract code which uploads the file under a guid from scratch each time. It would be better to upload the file under a content based hash so that we can avoid repeated uploads and storage of the same data.

Potentially the implementation here interacts with the way we could cache textract results in
aryn-ai/quickstart#3

[HenryL27]

can we close this now that we have caching?