feat: 完善 RecordStream 的协议协商、运行观测与文档说明

- 将 RecordStream 出站路径收敛为单 writer loop
  - 支持在 batch header 中 piggyback AckSeq，保留独立 ack 作为兼容回退
  - 增加 record stream 打开阶段能力协商，支持 mixed-version peer 自动降级
  - 补充 RecordSnapshot 与 diagnostics summary 的 record-plane 观测项
  - 增加 batch/ack/error frame、piggyback ack、barrier 等待拆分与 apply backlog 指标
  - 收紧 TransportConn detach 后的 runtime snapshot 语义
  - 补充 README 中的 RecordStream 语义、兼容行为与诊断快照说明
  - 补充相关单测与 race 回归验证

stream_control.go

@@ -20,6 +20,7 @@ type StreamOpenResponse struct {
 	DataID              uint64
 	Accepted            bool
 	TransportGeneration uint64
+	Metadata            StreamMetadata
 	Error               string
 }
 
@@ -95,6 +96,7 @@ func (c *ClientCommon) handleInboundStreamOpen(msg *Message) {
 		req.DataID = runtime.nextDataID()
 		resp.DataID = req.DataID
 	}
+	req.Metadata, resp.Metadata = negotiateRecordStreamOpenMetadata(req.Channel, req.Metadata)
 	stream := newStreamHandle(c.clientStopContextSnapshot(), runtime, scope, req, c.currentClientSessionEpoch(), nil, nil, 0, clientStreamCloseSender(c), clientStreamResetSender(c), clientStreamDataSender(c, c.currentClientSessionEpoch()), runtime.configSnapshot())
 	stream.setClientSnapshotOwner(c)
 	stream.setAddrSnapshot(c.clientStreamAddrSnapshot())
@@ -180,6 +182,7 @@ func (s *ServerCommon) handleInboundStreamOpen(msg *Message) {
 		req.DataID = runtime.nextDataID()
 		resp.DataID = req.DataID
 	}
+	req.Metadata, resp.Metadata = negotiateRecordStreamOpenMetadata(req.Channel, req.Metadata)
 	stream := newStreamHandle(logical.stopContextSnapshot(), runtime, scope, req, 0, logical, transport, streamTransportGeneration(logical, transport), serverStreamCloseSender(s, logical, transport), serverStreamResetSender(s, logical, transport), serverStreamDataSender(s, transport), runtime.configSnapshot())
 	if err := runtime.register(scope, stream); err != nil {
 		resp.Error = err.Error()