Add support for Gzip encoding

[ariofrio]

Some websites, like Wikipedia, send us a Gzip-encoded web page even if we don't ask for it. This usually happens when the page is very large. Try this to see what I mean:

curl -vI http://en.wikipedia.org/wiki/Spanish_language |& grep '^[<>]'

This patch helps us deal with those websites, and speeds things up, at least on Python 3.2 and up. Previous versions don't support GzipFile streaming, so we read the whole file to a StringIO before decoding with GzipFile.

[SimonSapin]

Hi! Thanks for contributing, and so sorry for letting this linger so long…

I took the liberty to rewrite this instead of taking your PR, in order to:

• Use the byte-based io.BytesIO instead of the Unicode-based io.StringIO or the ambiguous StringIO.StringIO
• When gzip streaming is not supported, just return a byte string. (As is done e.g. for data: URLs.) There is no use to returning a file-like object.
• Add some tests