Will close #12. For now this still has many problems and missing features.

• Bail early on failing tests
• Use reporter to output test results
• Fix test counts (not sure if they're correct atm, see Incorrect test count when nesting describe #10 too)
• There's a lot of unused code and layers of abstractions that we can do without
• setExpectedAssertionCount/incrementAssertionCount is tricky now that we have async. Somehow we need to scope that.
• beforeAll is called too often causing a failing test atm
• Create generic 'test/describe context' where assertion counts, isBeforeAllCalled, etc. can be stored. Only after this fix:
• Retry failed tests

I think the most straight forward fix for tracking assertion counts, is to create a text context and bind and pass that to the test as an 'expect' parameter. That way it would know which assertions to count and increment.

Alternatively, a way that doesn't change the API, could be to keep track of test source code lines and have expect find which test it is called in that way.

local function getCallerFunctionLine()
    local info = debug.getinfo(3, "S") -- stack level 3 == caller of `expect`
    return info and info.linedefined
end