FileReader: merge even when md5 differs.

Instead of bailing out if the md5 sum differs, just use the latter md5.
I think this should actually be a better strategy than recalculating the
md5 (and dealing with files that are not on disk).