Fix crash, false-positive in no-statement-after-end

[ninevra]

Fixes #315 by changing the way no-statement-after-end records CodePathSegments. The stack is used only to hold the in-progress segment in the outer CodePaths. When CodePaths begin or end, the current segment is pushed onto or popped from the stack. When CodePathSegments begin or end, the current segment is set or unset.

When in unreachable code at the end of a path, the current segment is therefore undefined, rather than erroneously taken from the next-outermost path.

Checks whether the current segment is set before operating on it.

(Interestingly, this rule doesn't catch statements after t.end() in unreachable code, e.g.:

throw new Error();
t.end();
1; // <-- does not trigger a report

This behavior predates this PR.)

While I believe this fixes the issue, I don't fully understand how this rule ought to handle its edge-cases nor why the codepath analysis system works this way. Please review carefully?

[novemberborn]

@sindresorhus could you take a look at this when you get some time? 😄

[sindresorhus]

I don't have a lot of experience with code path analysis, but the changes look ok from what I can tell.