extra: preserve hard_max after comparisons to zero

extra: preserve hard_max after comparisons to zero

John Levon reported that if you had code like:

int lx_cap_mapping[4];
int i = 63;

if (i)
	;
lx_cap_mapping[i] = 0;

The code should trigger a warning but it doesn't.

I initially thought the hard_max was getting lost in
merge_estates().  I changed that to say if we were merging an
impossible state with a possible one then we should keep the
hard max from the possible one.  It turns out that wasn't
the issue, but I sort of think it's the correct thing to do
anyway.

The real problem is that we don't preserve the hard_max after
comparisons to zero, whether it's impossible or not.  That is
handled in match_condition().  I changed that function to re-use
handle_comparison().  After that because the original "i" had a
hard_max then handle_comparison() sets the hard_max flag for
both the possible and the impossible sides.

Reported-by: John Levon <levon@movementarian.org>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>