Tests fail at high build concurrency for default openssh config

cargo test fails for me most of the time with --test-threads=24, even on a dual core machine, and ~30% of the time with $(nproc) == 12.

Random tests fail with: 'called Result::unwrap() on an Err value: Error { code: -43, msg: "Failed getting banner" }', at https://github.com/alexcrichton/ssh2-rs/blob/cab5f0fc9f217a577ad7614e5539e0d0315279b1/tests/all/main.rs#L27
-43 is LIBSSH2_ERROR_SOCKET_RECV.

If you enable debugging (cfg.define("LIBSSH2DEBUG", None); cfg.define("HAVE_GETTIMEOFDAY", None); and cripple the filtering) you can see that the connection is being closed.

Initially I thought this was a horrible thread safety bug (and spent ages on that), but it's probably just openssh's default MaxSessions 10, which means no more than 10 people can be logging in at a time, so, uh, probably a minor documentation note, instead of the cool bug I was hoping for?

Reproduce the failure with cargo test --test-threads=24 and inspect the referenced test at tests/all/main.rs, alongside the OpenSSH MaxSessions 10 behavior described in the issue. Determine where this concurrency limitation should be documented; done means the high-concurrency failure and relevant configuration or test guidance are clearly recorded.