bug#374: Info header line does not respect mouse-1-click-follows-link

[Drew Adams]

> > Whatever mouse-1 does on non-links, which is also whatever mouse-1
> > does elsewhere (e.g. non-header lines) when the variable is nil. In
> > most cases, it is what `mouse-set-point' does.
> 
> But mouse-set-point makes no sense on the header-line: there's no
> associated buffer position.

Wrong. You can click a non-link in the header-line or mode-line to set focus:
select its window/buffer. That is a primary use of the normal mouse-1 binding.

When mouse-1 follows links, you have to be careful where you click, to do that.
The bug is that for these screen areas, mouse-1 follows links regardless of the
option value.

> It's really like the mode-line, where
> mouse-1-click-follows-link is usually not taken into account either.

`mouse-1-click-follows-link' should also be taken into account in the mode-line,
for the same reason. Same bug or separate bug - take your pick.

Prior to the existence of option `mouse-1-click-follows-link', mouse-1 never
followed links - anywhere, ever. Users should still be able to obtain that
(superior) behavior. With that choice, users are able to click parts of Emacs at
nearly any location, to (typically) set focus.

IMO, the _default_ behavior of mouse-1 should be to not follow links. Failing
that change in default, users should at least have the _option_ to turn off link
following by mouse-1 everywhere. mouse-2 already follows links - there is no
reason to impose on _all users_ the redundancy of mouse-1 also doing that.

You cannot even use 0 as the value of the option in order to turn off link
following. That too is a (different) bug. The doc says that if the click time is
greater than the option value, then links are not followed: "The absolute
numeric value specifices the maximum duration of a "short click" in
milliseconds." If 0 is the max duration of a short click, then anything longer
than 0 should be treated the same as a long click.

Both 0 and nil should turn off link following by mouse-1 - *everywhere*. There
is no reason not to provide users with this pre-Emacs 22 behavior as an option.
Anything less is a regression.