During heavy parallel filesystem activity it was possible to Oops the
kernel. The reason is that read_cache_pages() could skip pages which
have already been inserted into the cache by another task.
Occasionally this may result in zero pages actually being sent, while
fuse_send_readpages() relies on at least one page being in the
request.
So check this corner case and just free the request instead of trying
to send it.