Don't update types of the outer scope when in an elseif conditional

[robchett]

This feels extreme for #9894 but I'm struggling to come up with a situation where the condition in an elseif can safely modify the outer scope unless the elseif could actually just be an if (the if branch was an early return).

[orklah]

Could you investigate the CI failure in Shepherd?

The fact that the only failure is in a test that is unrelated to if/else is both concerning and encouraging. Concerning because you'd expect that kind of feature to be covered by at least one test, but if it's not, it may means it's useless !

[robchett]

Looks like the Shepard failure is surfacing an error as the result of the bug fix

            $classNameEnd = strpos($function, '::');
            if ($classNameEnd !== false) {
                $className = substr($function, 0, $classNameEnd);
                if (!class_exists($className, false)) {
                    continue;
                }
            } elseif (!function_exists($function)) {
                continue;
            }

            yield "$function: " . json_encode($entry) => [$function, $entry];

The if doesn't verify that $function is a callable-string, and I'm not convinced they are callable-strings as the methods are not all static.
The usages of this method don't appear to require callable-strings anyway so I've corrected the return to simply string

[orklah]

Let's keep that one in mind, it's rare that removing code just works, but given we have no tests, any feedback will have the advantage of giving a useful test case