While using `telnetlib` I sometimes received unusually "large" messages (around 1Mb) from another process on the same machine, and was surprised `read_until` took more than a second. After instrumenting I discovered such messages were received at roughly 500kbyte/s.

I think this low throughput comes from two implementation details:

 - `Telnet.fill_rawq` is calling `self.sock.recv(50)`, whereas 4096 is now recommended
 - the `Telnet.process_rawq` method is transferring from raw queue to cooked queue by appending byte per byte.

For the latter, transferring by slices looks much faster (I'm measuring at least 5x). I'm preparing a PR.