fix race condition with client and serial endpoints

When a node was being closed, the read routine wasn't shut down
properly, causing Read() to be called again and again, until an error.
Since multibuffer.Next() is called just before Read(), the same buffer
was being accessed by multiple routines.

endpoint_client.go

@@ -161,6 +161,7 @@ func (t *endpointClient) runInner() error {
 				select {
 				case t.read <- buf[:n]:
 				case <-t.ctx.Done():
+					return errTerminated
 				}
 			}
 		}()

endpoint_serial.go

@@ -135,6 +135,7 @@ func (t *endpointSerial) runInner() error {
 				select {
 				case t.read <- buf[:n]:
 				case <-t.ctx.Done():
+					return errTerminated
 				}
 			}
 		}()