Accept output buffer in Response::Body#readpartial

[janko]

This makes it possible to use a HTTP::Response::Body object as a source argument in IO.copy_stream, allowing easy streaming. It also allows it to be used in the request body for another request (as HTTP::Request::Body::Writer uses IO.copy_stream), should anyone want that.

response = HTTP.get("http://example.com/download")
IO.copy_stream(response.body, destination)

Since IO.copy_stream uses an output buffer, it makes retrieving the response body much more memory efficient, because we're able to deallocate retrieved chunks. I tested with a 50MB response body, and streaming with HTTP::Response::Body#each allocated about 52MB of strings, whereas the IO.copy_stream version allocated only 3MB.

IO.copy_stream requires #readpartial to have IO#readpartial semantics, which among others includes raising and EOFError when end of file was reached. Since IO.copy_stream always uses an output buffer, the exception is raised only when output buffer is specified.

[janko]

This one needs JRuby 9.2.0.0 to pass the tests (which has the fix for jruby/jruby#5167), but it's currently failing because of #475.

[janko]

Now that I think about this more, this is probably an overkill. I don't like the idea of HTTP::Response::Body#readpartial partially mimicking IO#readpartial, it doesn't need to act as an IO object. People can just clear yielded chunks themselves if they want to save on memory.