df: use system-wide setting for our local config of require-confirmed

We were saving the initial setting and re-using it for the life of the
channel, but I think it makes sense to update to using whatever the
system setting is at restart?

We're leaving a record of what was originally done at start for now,
either so splicing can reuse or future cleanup.

Changelog-Changed: On RBF, v2 opens use the current system-wide setting for `require-confirmed-inputs`, instead of the setting from when channel was initialized

lightningd/dual_open_control.c

@@ -4170,7 +4170,7 @@ bool peer_restart_dualopend(struct peer *peer,
 				      amount_sat_zero(inflight->lease_amt) ?
 					      NULL : &inflight->lease_amt,
 				      channel->type,
-				      channel->req_confirmed_ins[LOCAL],
+				      peer->ld->config.require_confirmed_inputs,
 				      channel->req_confirmed_ins[REMOTE]);
 
 	subd_send_msg(channel->owner, take(msg));

tests/test_opening.py

@@ -2408,13 +2408,10 @@ def _no_utxo_response(r):
                            min_witness_weight=110,
                            excess_as_change=True)['psbt']
 
-    # Attempt bump, fail. L2 should remember required-confirmed-inputs
-    # from original channel negotiation, despite node-wide setting
-    # being flagged off
+    # Attempt bump, L2 uses the system-wide setting now.
     l1.rpc.connect(l2.info['id'], 'localhost', l2.port)
     bump = l1.rpc.openchannel_bump(res['channel_id'], amount, psbt)
-    with pytest.raises(RpcError, match=r'Input .* is not confirmed'):
-        l1.rpc.openchannel_update(res['channel_id'], bump['psbt'])
+    l1.rpc.openchannel_update(res['channel_id'], bump['psbt'])
 
     _verify_utxos(l1, utxo_lookups)