According to the documentation the destination IO only needs to respond
to #write. However, the downloader breaks this by calling #flush.
We fix this by calling #flush only if the destination IO responds to
that method. We also add a test to enforce that the destination IO only
needs to respond to #write.
Not all IO objects know how to #rewind themselves. For example, Ruby
pipes (returned by `IO.pipe`) do implement #rewind, but they will throw
an error if you try to call it.
rd, wr = IO.pipe
wr.rewind # Errno::ESPIPE: Illegal seek
But we don't need to rewind and overwrite the IO object if we didn't get
the Ranged response we expected, we could instead wait out the content
that has already been downloaded, and start appending again once we
reached where we left off. This is what this commit does.
Janko Marohnić
Steve Bazyl
Steven Bazyl