Test strictness signatures of base for regressions

We frequently accidentally find (#18187 (closed)) or introduce (!3110 (comment 272544)) regressions in base and friends.

These changes should not be accidental and caught by CI. Therefore I propose to test strictness signatures of functions exported from base recorded in interface files against a Golden Master, like we do for smaller tests in stranal.

We can possibly extend that in the future to also compare other kinds of analysis info, like CPR or Arity.

Compared to using something like StrictCheck, we won't have to come up with a specification for every function in base, which scales O(n) in the number of functions. The constants of writing such a test that isn't too noisy are arguably high, though.