Preserve original uploads in processing pipeline

[eyeseast]

The current processing pipeline deletes the original uploaded file. https://github.com/MuckRock/documentcloud/blob/master/documentcloud/documents/processing/document_conversion/main.py#L178-L179

We should keep these as a source-of-truth (possibly in cheaper storage).

[allanlasser]

This only applies to non-PDF file types.

[allanlasser]

Some questions that came up during triage:

• What % of uploads are non-PDF?
• Can we expose the URL or original extension through the API?

[mitchelljkotler]

Some questions that came up during triage:

* What % of uploads are non-PDF?

I just checked, 91% of documents are originally PDF