StandardConstraintResolver interface needs to return errors

[rodaine]

The current StandardConstraintResolver interface does not support returning errors, preferring the implementation to fail silently if it cannot resolve the constraints. But in practice, resolving the extension is fallible (very apparent in #57). This will be a breaking change to the interface and a behavioral change to the library (instead of failing silently and not producing a compilation error, the validator would return a compilation error caused by the resolver's error).