fix race condition if during reconfiguration two cache managers try to delete old inactive entries: one of them removes a entry just locked by other manager from the queue and the rbtree as long inactive entry, causes the latter manager to segfault leaving cache mutex locked, the bug has been introduced in r3727

# Copyright (C) Igor Sysoev


cat << END >> $NGX_AUTO_CONFIG_H

#ifndef $have
#define $have  1
#endif

END