Move the check for legacy/UTF character conversion on incoming URLs from

Move the check for legacy/UTF character conversion on incoming URLs from
Title::newFromURL into WebRequest itself. Should now work on all get params,
so essentially the full URL will be converted.

(The check is not done if the referer matches the canonical server, as before.)

Fixes problem with latin-1 typed URLs introduced by the stricter UTF-8
validation checks running before the conversion check.

One minor glitch; the canonical URL redirection no longer picks up on a
charset conversion. However it's broken anyway and doesn't pick up eg
non-canonical URLs using ?title= instead of / or other such things. Needs
to be improved...