Close multiplexer when context is cancelled

Multiplexer continues to create rpc connections even when
the context which is passed to the underlying rpc connections
is cancelled by the server.

This was causing hashicorp#4413 - when a SIGHUP causes everything to reload,
it uses context to cancel the underlying http/rpc connections
so that they may come up with the new configuration.
The multiplexer was not being cancelled properly so it would
continue to create rpc connections and constantly fail,
causing communication issues with other nomad agents.

Fixes hashicorp#4413

nomad/rpc.go

@@ -208,6 +208,11 @@ func (s *Server) handleMultiplex(ctx context.Context, conn net.Conn, rpcCtx *RPC
 	s.setupRpcServer(rpcServer, rpcCtx)
 
 	for {
+		// stop handling connections if context was cancelled
+		if ctx.Err() != nil {
+			return
+		}
+
 		sub, err := server.Accept()
 		if err != nil {
 			if err != io.EOF {
@@ -311,6 +316,11 @@ func (s *Server) handleMultiplexV2(ctx context.Context, conn net.Conn, rpcCtx *R
 	s.setupRpcServer(rpcServer, rpcCtx)
 
 	for {
+		// stop handling connections if context was cancelled
+		if ctx.Err() != nil {
+			return
+		}
+
 		// Accept a new stream
 		sub, err := server.Accept()
 		if err != nil {