Premature destroy in NAT detection

[nanangizz]

Reported crash in NAT detection with call stack trace like:

Exception Type:  EXC_BAD_ACCESS (SIGSEGV)

0   grp_lock_acquire + 68 (lock.c:297)
1   pj_grp_lock_acquire + 24 (lock.c:486)
2   on_sess_timer + 160 (nat_detect.c:915)
3   pj_timer_heap_poll + 616 (timer.c:912)

A possible cause is a race condition between timer and socket which can cause NAT detect session to be prematurely destroyed.