Update checker for #248 substitution decision

[cpovirk]

I have a note that I saw what looked like a related problem in build 017dc798-8948-4391-8614-d59a2b2f7c43. It happened when Map.merge had an unspecified-nullness parameter for V value under J2CL and some Guava code naturally had @NonNull V value. A non-null subclass parameter should be considered valid when overriding a method with an unspecified superclass parameter.

I also have a note that either I thought it might be related to #193 or just that I figured #193 was another thing that I should look into at the same time.

The two do seem plausibly related and might even be fixed by my internal CL 664839751, which implements at least some of the changes of jspecify/jspecify#248 (hopefully all of them).