netfilter: nft_compat: remove flush counter optimization

WARNING: CPU: 1 PID: 16059 at lib/refcount.c:31 refcount_warn_saturate+0xdf/0xf
[..]
 __nft_mt_tg_destroy+0x42/0x50 [nft_compat]
 nft_target_destroy+0x63/0x80 [nft_compat]
 nf_tables_expr_destroy+0x1b/0x30 [nf_tables]
 nf_tables_rule_destroy+0x3a/0x70 [nf_tables]
 nf_tables_exit_net+0x186/0x3d0 [nf_tables]

Happens when a compat expr is destoyed from abort path.
There is no functional impact; after this work queue is flushed
unconditionally if its pending.

This removes the waitcount optimization.  Test of repeated
iptables-restore of a ~60k kubernetes ruleset doesn't indicate
a slowdown.  In case the counter is needed after all for some workloads
we can revert this and increment the refcount for the
!= NFT_PREPARE_TRANS case to avoid the increment/decrement imbalance.

While at it, also flush for match case, this was an oversight
in the original patch.

Fixes: ffe8923f ("netfilter: nft_compat: make sure xtables destructors have run")
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>