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.