Test sup command

[mingan]

This PR builds on top of #134. It adds tests for sup commad as a sanity check of the existing functionality. In order to accomplish it I had to adjust the code in few places:

1. I moved most of the code from main to runSupfile so that it can be run from tests
2. Most flags are parsed into a struct so that tests don't depend on global vars (and it's a struct instead of eight separate variables)
3. To keep the test output clean the error output stream is passed around explicitly and set to ioutil.Discard in tests
4. SSHClient got new fields so that I could remove the global variable initAuthMethodOnce preventing running the command multiple times (aka when running tests)

The tests which need to spin up mock SSH servers, capture the requests and always respond with success. I barely understand this part and it's put together from different examples around the web.

I'm doing this as a preparation for tackling #130.

[mingan]

Updated the test setup to get it running on Travis CI.

[VojtechVitek]

This looks awesome.

But I'm too busy and don't have time for a thorough review right now. I'll come back to this eventually

[VojtechVitek]

Can you explain in high level what the provided tests do exactly?

[mingan]

Sure, the basic idea is to move the most of the main function to a separate function. main now just gathers flags, sets up dependencies and calls this function. Tests do the same. Normal de-coupling from flags, global vars etc.

Each test spins up mock a set of SSH servers which only record command and provide them back for assertions, they don't do anything else. The server addresses are written into an SSH config file in a tmp directory alongside generated key and cert for the SSH servers, public and private key pair of the client.

Every test contains YAML string with the Supfile being tested. The test runs this Supfile from parsing YAML, to setting up App and actually performing the commands. matcher_test.go provides convenience functions for testing that commands were sent to the defined servers and can test individual exports or commands. It's based on comparing the requests sent to the servers as strings, nothing more.

The tests cover basic things like missing networks, commands, targets etc., simple command, sequence of commands, a target, --onlyHosts, --exceptHosts, hosts specified via inventory, default SUP_* variables, setting global, network and command line variables as well as custom file name.

The code for the mock servers is horrible. I assume it can be improved but I just stitched it together from different examples and by trial and error.

I'm happy to answer any more questions.