Add code to the pNFS client so that it shuts down a TCP connection to a
DS when a renew operation (actually just a Sequence Op) fails due to ENXIO.
Not particularily critical, but without this patch the client just does
a TCP reconnect attempt every 30sec.
Also, make the call to nfscl_cancelreqs() unconditional, since it checks
for NFSCLDS_SAMECONN and it does so with the lock held.