Measure true coverage (of mcpyrate tests)

[Technologicat]

As of 8b2b47c, coverage measurement includes code covered not only by tests, but also by demos.

Demos don't necessarily test anything other than that they don't crash, so the coverage report is overly optimistic.

We need a flag in runtests.py for whether it should run tests and demos (for testing), or tests only (for coverage measurement).

Results are likely horrifying at this point, but an honest result is better than an overly optimistic estimate.

Despite the abysmal test coverage, mcpyrate is actually working pretty well, it can run all of unpythonic just fine. It's only the automated tests that are lacking (see #2). Obviously we need a lot more tests, but producing an accurate coverage result is a good first step so we can actually see what needs to be done.