…NK_SIZE

MediaIoBaseDownload hangs forever when object to download is smaller than DEFAULT_CHUNK_SIZE

It turns out that the same code in `http.py` is used for processing both (1)
the initial response and redirect and (2) the replies containing actual bytes
for download. This means that the new exception we added for (2) leads to
problems in (1).

Remove a newly-added exception.

Currently, if we have a top-level methodPath of the form `path:methodName`,
we'll pass this to `urlparse.urljoin`, where this is interpreted as
`scheme:netloc`. In cases other than media upload, this is **not** what we
want.

This fixes the issue by adding our own custom `urljoin`, which detects and
handles this case. We also add a test.

Fixes #40.

Work toward Python 3.

Run modernize over the codebase (work toward Python 3).