Growing i_size in hugetlbfs_prefault() isn't right - if we run out of pages,
do_mmap_pgoff() will chop the partially-instantiated pages off again.

So update i_size in hugetlbfs_file_mmap() if the whole mmap attempt was
successful.