[gold][patch] Reduce heap usage for string merge sections

[Cary Coutant]

> There may be a minor win available by reserving the size of the
> merged_strings_list vector for a large input section.

Good idea -- do you have a suggestion for estimating that? I could
just divide the size of the section by a presumed average string
length. The average string length might depend heavily on the source
language and what the section is, though. (Although I suppose a
conservative estimate, based on a large average string length, is
better than doing nothing.)

-cary