bowling: propose tests for err semantics

[petertseng]

Specifically: if a call to roll() is made that results in an error,
the game state should not be altered, so that future calls to
roll() will proceed as if the erroneous roll never happened.

This is only a proposal because there was doubt in #213 about whether
they should be added.

Since it was suggested, after seeing it in my Exercism submission, that
I open an issue for it, I will, because I can let code speak to
accompany my words.

Right now there's no good way to express this kind of semantics in
x-common, but maybe a general discussion on it would be OK to brainstorm
some ways to do it.

TODO:

• The current example solution needs to be fixed.
• It is possible we can add more tests of this type - basically any time
  we expect an is_err() from a roll.

I would only want to put in the work to proceed with these if we decide
we do want these kinds of tests.

[IanWhitney]

I still think this adds new behavior to the exercise that I don't think is necessary. It gets into the territory of "If I were to implement a real bowling scorer", and I think exercises shouldn't worry about what 'real world' solutions; instead they should be focused on a single thing.

[petertseng]

It can sometimes be hard to know where to draw the line. We have seen before at #178 (comment)

But if people are using these exercises to learn how to write Rust and the exercises encourage smelly Rust code, then won't most of the students go off and write smelly Rust code in production?

I think currently we draw the line at essential behaviour, and we can't really say that this behaviour is essential to scoring a bowling game.

Another comfort we can take is that even though the exercise doesn't enforce the kind of discipline in error handling that we would be asking for in this test, at least it is not encouraging smelly code. We're leaving it open. So the stance we take is not inconsistent with the above quote.

I think if someone is very interested in error handling, they should help us implement the https://github.com/exercism/x-common/tree/master/exercises/error-handling exercise... where everything discussed here would suddenly become essential to the problem... since it's the entire problem.