Better headers

[ml31415]

Currently, the original case sensitive header fields are only preserved for the first lines with matching header fields. This is nothing too bad, as the RFC only demands a correct case-insensitive treatment, but it's not nice as well.

This PR now stores all case-sensitive header fields along with the header values. While at the same time it sticks with the principle of handling a single header field without any duplication, as default case with nearly the handling speed of an ordinary python dictionary.

Iterating over all header lines with their original cases is now the default behaviour of Header.items() and Headers.iteroriginal() is deprecated. So as a result, Headers can now be formatted nearly as the plain HTTP hext received over the wire.

>>> h = Headers(asdf="ddd", ASDF="fff", AsDf="asdfasdf")
>>> print(h)
asdf: ddd
ASDF: fff
AsDf: asdfasdf

[cyberw]

Cool stuff. Just a question: Does this make us more similar to requests or less?

[ml31415]

More. But we're quite similar anyways, as I had contributed a similar headers implementation as our current one to urllib3 ~10 years ago.
urllib3/urllib3@5e73e72

For requests compatibility, I suggest to add some thin front-end to UserAgent, so that it can serve as a drop-in replacement for requests for simple use cases. I had an attempt to use geventhttpclient as backend for requests instead of urllib3, but that didn't achieve much, as requests wasted so much time on top, that it completely defeated the original goal.