There were situations, when downstream handshake stuck and stayed in
mtp_dc_pool `pending` field forever.
With this change if downstream can't finish handshake in 8s, it will crash

* fix epmd detection on CentOS
* get rid of awk dependency

Also, replace broken IP lookup service with some others

Because of tls1.3 fixes

By providing key_share and upported_versions extensions in ServerHello

See rfc8446#section-5.1

* pretends being TLS1.3 + http2
* mtp_handler state-machine refactored

See https://github.com/erlang/docker-erlang-otp/issues/235

Buffer was not saved if crypto codec returns `incomplete`

* Get rid of per-codec internal read buffer