Give users more leeway to reject edge cases in different stages

[nlsandler]

There are some error cases that it would be reasonable to catch in either the parser or the type checker. As @ysono points out, some tests of intermediate stages will fail if you decide to handle these error cases in a different stage than the test suite expects. It should be possible to make reasonable choices that diverge slightly from what's in the book without getting test failures.

Possible fixes:

1. Add a --skip-temporarily-valid option, which tells the compiler to skip any test cases that are invalid but should be accepted by the current stage. (E.g. if you're running --stage parse, just don't run the tests in invalid_types. Would need to think on this more to make sure it won't break anything/undermine test coverage. Downside: doesn't help if the user decides to handle a case later than the test suite expects, only if they decide to handle it earlier.
2. Provide a configuration file where users can specify the expected failure stage for individual tests. More flexible but also more complicated.