fix: implement InvokeCallback.operationFail() for rocketmq-remoting 5.5.0 compatibility

[Copilot]

PR #334 upgraded rocketmq-remoting 5.1.0 → 5.5.0 alongside the fastjson2 migration. The 5.5.0 release added operationFail(Throwable) to InvokeCallback and NettyRemotingClient.invokeAsync() now calls it directly (bypassing operationComplete) when connections fail — null channel, inactive channel, connection timeout. DLedger's lambda callbacks only implemented operationComplete, so operationFail was a default no-op, leaving CompletableFutures permanently incomplete on any connection failure.

Impact: Leader election, heartbeat, vote, push, and snapshot operations would hang for 3 s before TimeoutException, causing cascading failures in 13 CI tests across LeaderElectorTest, AppendAndPushTest, BatchPushTest, CommitIndexTest, and SnapshotManagerTest.

Changes

• DLedgerRpcNettyService — replace all 6 lambda InvokeCallback instances (heartBeat, vote, append, push, installSnapshot, leadershipTransfer) with anonymous classes that implement operationFail(), immediately completing the future with NETWORK_ERROR so callers can retry without waiting for the 3 s .get() timeout:

// Before — operationFail is a no-op; future never completes on connection failure
remotingClient.invokeAsync(addr, request, 3000, responseFuture -> {
    RemotingCommand responseCommand = responseFuture.getResponseCommand();
    if (responseCommand != null) { ... }
    else { future.complete(NETWORK_ERROR); }
});

// After — connection failures surface immediately
remotingClient.invokeAsync(addr, request, 3000, new InvokeCallback() {
    @Override
    public void operationComplete(ResponseFuture responseFuture) {
        RemotingCommand responseCommand = responseFuture.getResponseCommand();
        if (responseCommand != null) { ... }
        else { future.complete(NETWORK_ERROR); }
    }

    @Override
    public void operationFail(Throwable throwable) {
        future.complete(NETWORK_ERROR);  // unblocks caller immediately
    }
});