fix: restore v1 intercept parity on master

Restore the previously merged macOS VPN DNS post-settle refresh,
unchanged-exemption suppression, forced pf state flush, and self-upgrade
test after they were removed by the firewall-mode merge.

Also keep Windows NRPT pointed at loopback when the listener is configured
on a wildcard address.

cmd/cli/dns_intercept_darwin.go

@@ -1233,6 +1233,10 @@ func (p *prog) pfStabilizationLoop(ctx context.Context, stableRequired time.Dura
 			p.pfStabilizing.Store(false)
 			mainLog.Load().Info().Msgf("DNS intercept: pf stable for %s — restoring anchor rules", stableRequired)
 			p.ensurePFAnchorActive()
+			routes, domainlessServers, exemptions := p.refreshDNSAfterVPNSettle("pf_stabilized")
+			if routes == 0 && domainlessServers == 0 && exemptions == 0 {
+				p.scheduleDNSAfterVPNSettleRefresh("pf_stabilized_followup", pfAnchorRecheckDelayLong)
+			}
 			p.pfLastRestoreTime.Store(time.Now().UnixMilli())
 			return
 		}
@@ -1459,6 +1463,15 @@ func isResourceExhaustion(err error, output []byte) bool {
 		strings.Contains(msg, "cannot allocate memory")
 }
 
+func (p *prog) scheduleDNSAfterVPNSettleRefresh(reason string, delay time.Duration) {
+	time.AfterFunc(delay, func() {
+		if p.dnsInterceptState == nil {
+			return
+		}
+		p.refreshDNSAfterVPNSettle(reason)
+	})
+}
+
 // pfWatchdog periodically checks that our pf anchor is still active.
 // Other programs (e.g., Windscribe desktop app, macOS configd) can replace
 // scheduleDelayedRechecks schedules delayed re-checks after a network change event.
@@ -1826,7 +1839,14 @@ func (p *prog) forceReloadPFMainRuleset() {
 		mainLog.Load().Error().Err(err).Msgf("DNS intercept: force reload — failed to load anchor (output: %s)", strings.TrimSpace(string(out)))
 	}
 
-	// Reset upstream transports — pf reload flushes state table, killing DoH connections.
+	// Flush stale rdr/reply states after the forced ruleset + anchor reload.
+	// Without this, macOS can keep using pre-reload state and try to send
+	// redirected DNS replies directly from loopback to tunnel client addresses
+	// (for example, 127.0.0.1:<listener> -> 100.64.0.0/10), which fails with
+	// "sendmsg: can't assign requested address".
+	flushPFStates()
+
+	// Reset upstream transports — pf reload/state flush kills existing DoH connections.
 	p.resetUpstreamTransports()
 
 	mainLog.Load().Info().Msg("DNS intercept: force reload — pf ruleset and anchor reloaded successfully")