Passing chunk_size=None to iter_content() works fine for chunk-encoded
responses, but still leaves us buffering the entire response when the
server just specifies a Content-Length.
Now, stream the response 256k at a time.
Why 256k? It's large enough that you won't be in a tight loop doing tiny
reads, but small enough that it shouldn't be any great memory concern.
If anyone finds that it's problematic, they can always monkey-patch
quantconnect.api.DOWNLOAD_CHUNK_SIZE.
Previously, the entire response would be read and buffered in memory
before any of it was written to disk.
Now, stream the response and write down data as it's received.
Tim Burke