This updates MIMEAccept to mimic old behaviour better so that downstream
projects are not as broken.

The goal is not to provide the exact same behaviour, and thus MIMEAccept
still uses the new Accept header parsing and matching, however it is
much closer to the old API because it no longer raises on an invalid
header vs what was shipped in WebOb 1.8.0.

(cherry picked from commit 3fa3311)

The main change here is to simplify the handling of duplicate ranges in
the header. The RFC does not specify what to do when the same range
appears in the header more than once, possibly with different qvalues
and giving conflicting information. Previously, we chose which of the
possibly conflicting versions of the range to use for matching by first
checking if there was one with q=0, which would take priority; if one
with q=0 was not found, we chose the one with the highest qvalue.

Having given it a lot more thought, it now seems clear that this kind of
special handling is not necessary -- the simplest way to handle it is to
take the first appearance of the range in the header (from the left) and
simple use that for matching.