bugfix: Clear .hello_acc in mtp_handler when no longer needed

Before fix accumulator was never cleaned, taking memory.
Clear the accumulator on transition to tunnel stage (handshake complete)
and on transition to fronting stage (data already forwarded to front_sock).

Author: seriyps

src/mtp_handler.erl

@@ -427,7 +427,8 @@ parse_upstream_data(<<Header:64/binary, Rest/binary>>,
                         dc_id = {RealDcId, Pool},
                         codec = Codec,
                         policy_state = PState,
-                        stage = tunnel},
+                        stage = tunnel,
+                        hello_acc = <<>>},
                 hibernate));
         {error, Reason} when is_atom(Reason) ->
             mtp_metric:count_inc([?APP, protocol_error, total], 1, #{labels => [Listener, Reason]}),
@@ -538,7 +539,7 @@ do_front(SniDomain, Config, Data, Ip, Listener,
                 ok = gen_tcp:send(FrontSock, Data),
                 ok = Transport:setopts(Sock, [{active, once}]),
                 ?LOG_INFO("Domain fronting to ~s:~p for SNI ~s", [Host, Port, SniDomain]),
-                {ok, S#state{stage = fronting, front_sock = FrontSock}};
+                {ok, S#state{stage = fronting, front_sock = FrontSock, hello_acc = <<>>}};
             {error, Reason} ->
                 ?LOG_WARNING("Domain fronting connect to ~s:~p failed: ~p",
                              [Host, Port, Reason]),