The workhorse _perform_request eagerly parses request data either as JSON or as CSV. This is likely fine for most users, but for users who would rather not parse the data with Python, it would be handy to have a way to opt-in to receive response.content (or a streaming response.raw). In my use case, Python is passing the data from an API to S3, so I don't need it to parse anything.

A related issue is that the parsing done by the JSON parser and the CSV parser returns different iterators. The first returns an iterator over dictionaries, while the second returns an iterator over lists (with the header inlined, which runs into #107). It would be nice if this was consistent, but not a deal breaker.