Re: gold patch: fix global symbols in merge sections

["Cary Coutant" <ccoutant at google dot com>]

>> several uses of -1U that did not properly extend to 64 bits.
>
> If you use -1 then you won't get that problem.

And that's what we use when the type is signed -- there are places in
gold that use -1 for signed types and I left those alone. But if the
type is unsigned, using -1 in comparisons produces warnings (we
compile gold with -Wall -Werror) and just seems dangerous. Using a
properly typed and named constant for this special "invalid" value is
cleaner, in my opinion.

-cary