BUG/MINIR: h1: Fix doc of 'accept-unsafe-...-request' about URI parsing

BUG/MINIR: h1: Fix doc of 'accept-unsafe-...-request' about URI parsing

The description of tests performed on the URI in H1 when
'accept-unsafe-violations-in-http-request' option is wrong. It states that
only characters below 32 and 127 are blocked when this option is set,
suggesting that otherwise, when it is not set, all invalid characters in the
URI, according to the RFC3986, are blocked.

But in fact, it is not true. By default all character below 32 and above 127
are blocked. And when 'accept-unsafe-violations-in-http-request' option is
set, characters above 127 (excluded) are accepted. But characters in
(33..126) are never checked, independently of this option.

This patch should fix the issue #2906. It should be backported as far as
3.0. For older versions, the docuementation could also be clarified because
this part is not really clear.

Note the request URI validation is still under discution because invalid
characters in (33.126) are never checked and some users request a stricter
parsing.