Reduce test_exercise duplication + fix reporter statefulness

[filipsch]

• There is a test_exercise function in sqlwhat and shellwhat, but not in protowhat. Seems like test_exercise should be moved there, with a smart way of including the language-specific SCT context and state

• The reporter depends on a singleton-like failed_test variable, that it uses to build up a payload. It is not required. Thoughts by @machow about this:

I regret never finishing cleaning up Reporter (and taking out failed_test). It would super clean to have reporter just hold options for reporting outcomes, and then rather than carrying around the backend output with it, test_exercise could pass that at the end. See [here](see https://github.com/datacamp/protowhat/blob/fs/test-not/protowhat/Reporter.py#L57-L58). Something like

def fake_test_exercise(output):
    reporter = Reporter()
    try: test_something('bleh')
    except TestFail as e:
        return reporter.build_payload(e)
    return reporter.build_payload(output)

[machow]

For some added clarification, reporter.failed_test is leftover from before SCTs would mark this attribute as true, rather than raising an error. Now, it seems like something like test_exercise should fail from two places: running the SCTs raised TestError, or Reporter judges the backends output to contain an error.

In library's like pythonwhat, I think there are a few places that set failed_test (but if changed to raise an error instead, should be okay; also protowhat isn't used there).