fix: enforce per-channel MQTT connection state guard

[wy471x]

Make sure that:

• You have read the contribution guidelines.
• You submit test cases (unit or integration tests) that back your changes.
• Your local test passed ./mvnw clean install -Dmaven.javadoc.skip=true.

Summary

Changes:

1. MessageType.java — connection state moved from a per-instance volatile boolean (dead: MqttFactory builds a fresh MessageType per inbound message, so the state set by Connect.connect() never propagated to other handlers) to a per-channel AttributeKey<Boolean>; isConnected(Channel) / setConnected(Channel, boolean) now read/write the channel attribute.
2. Publish.java, Unsubscribe.java, PingReq.java — guards inverted to if (!isConnected(ctx.channel())); like Subscribe, the channel is now closed (FIRE_EXCEPTION_ON_FAILURE) instead of silently dropping the packet.
3. Subscribe.java — inverted the dead if (isConnected()) guard.
4. Connect.java — duplicate CONNECT on the same channel is rejected by closing the connection without sending a CONNACK ([MQTT-3.1.0-2]).

Test Cases:

1. ConnectTest — migrated from Mockito mocks to a real EmbeddedChannel (per-channel attributes require a real channel); added duplicateConnectIsRejected (second CONNECT receives no CONNACK and closes the channel).
2. PublishTest — added publishBeforeConnectClosesChannel and the end-to-end publishAfterConnectOnSameChannelIsAccepted (real Connect.connect() followed by Publish.publish() on the same channel).
3. New PingReqTest — pingReqBeforeConnectClosesChannel, pingReqAfterConnectSendsPingResp.
4. New UnsubscribeTest — unsubscribeBeforeConnectClosesChannel.

Verification

• shenyu-protocol-mqtt module: 13 tests passed (ConnectTest 5, PublishTest 5, PingReqTest 2, UnsubscribeTest 1), JDK 21.
• Checkstyle and RAT (license header) checks passed.

close #6638

[Aias00]

The duplicate-CONNECT path looks like it needs cleanup before closing the channel. A successful first CONNECT registers the channel in , but this new early return only calls . I only see repository cleanup in the explicit DISCONNECT flow, and there does not appear to be centralized \ cleanup for MQTT channels, so a client that sends CONNECT twice can leave a closed channel referenced in . Could you remove the channel from \ before closing here, or add centralized channel-inactive cleanup for all server/client close paths?

[Aias00]

Correction to my previous review comment, where inline code formatting was stripped by my local shell:

The duplicate-CONNECT path looks like it needs cleanup before closing the channel. A successful first CONNECT registers the channel in ChannelRepository, but this new early return only calls ctx.close(). I only see repository cleanup in the explicit DISCONNECT flow, and there does not appear to be centralized channelInactive cleanup for MQTT channels, so a client that sends CONNECT twice can leave a closed channel referenced in ChannelRepository. Could you remove the channel from ChannelRepository before closing here, or add centralized channel-inactive cleanup for all server/client close paths?