This PR improves our handling of HTTP message body encodings:
- The unaltered message body is now accessible as `.raw_content`
- The "content-encoding"-decoded content (i.e. gzip removed) content
is not `.content`, as this is what we want in 99% of the cases.
- `.text` now provides the "content-encoding"-decoded and then
"content-type charset"-decoded message body.
- The decoded values for `.content` and `.text` are cached,
so that repeated access and `x.text = x.text` is cheap.
- The `decoded()` decorator is now deprecated, as we can now just use
`.content`. Similarly `HTTPMessage.get_decoded_content()` is
deprecated.
- The canonical source for :method, :scheme and :path are the .method, .scheme
and .path attributes on the request object.
- These pseudo-headers are stripped after reading the request, and re-inserted
just before sending.
- The :authority header remains, and should be handled analagously to the Host
header in HTTP1 with respect to display and user interaction.
This commit introduces MultiDict, a multi-dictionary similar to
ODict, but with improved semantics (as in the Headers class).
MultiDict fixes a few issues that were present in the Request/Response
API. In particular, `request.cookies["foo"] = "bar"` has previously been a
no-op, as the cookies property returned a mutable _copy_ of the cookies.
Currently the path_components and query setters of the Request object
use the url setter under the hood. The url setter updates all parts of
the URL including the host. If the host header and the host in the
request URL are different (as is common when making HTTPS requests)
then the host header will be updated to the value in the URL as a
result of modifying the path.
This change fixes this problem by modifying the query and
path_components setters to not use the url setter and instead directly
update the path field.
