socket: Use modulo-N for SO_REUSEPORT socket locating

socket: Use modulo-N for SO_REUSEPORT socket locating

This greatly reduces TCP listen socket completion queue pool token
contention thus saves more cpu time.

When the testing box (i7-2600, broadcom 5719) is doing 335Kconns/s,
TCP listen socket completion queue token contention is reduced from
20K/s to 500/s and the idle time on each hyperthread is increased by
~5%.