There were two loops, one to skip entries before the requested offset,
and one to populate entries. Merge them and use the return value of
pfs_iterate() to decide whether to set eofflag.

As a side effect, this fixes a small bug: the first loop would decrement
offset to zero, and the second loop would increment it again.
However, offset was used to set d_off, and we should include the
starting offset there.