Feature request: read data from in-memory buffer

[philbooth]

My use-case is that we have a bunch of Parquet files in S3 I'm operating over in batches. While it works fine to download things to the local file system before reading and then deleting them after I'm done, it would be nicer if I could cut out the file system completely and pass the reader object a Buffer instance.

(and apologies if this already exists but I just didn't spot it in the docs)