netfilter: ecache: don't use nf_conn spinlock

netfilter: ecache: don't use nf_conn spinlock

For updating eache missed value we can use cmpxchg.
This also avoids need to disable BH.

kernel robot reported build failure on v1 because not all arches support
cmpxchg for u16, so extend this to u32.

This doesn't increase struct size, existing padding is used.

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>