assert that eval() returns ok where it should

[gsauthof]

This detects failed attempts where the stack looks as expected but
eval still returns the wrong result.

Also, this should point out some implementation errors more clearly
because you see a specific error result and not just the mismatching
stack content. Especially when there are multiple eval() calls in a row.

[petertseng]

I think I'm for this. I would like to see the #![allow(unused_must_use)] on line1 removed too if this is going to happen, please

[gsauthof]

Good point. The first line with that warning suppressing attribute is gone now.

(And, as expected, because of the new asserts no warnings are emitted.)

[petertseng]

Thanks. I am for this and I agree with the reasoning.

Once 48 hours have passed since it was first opened (<12 hours from now) I'll merge if nobody else has any else to say.

It is interesting that we did not use the same reasoning in bowling (#213). Should we?