special handling for Completion/NormalCompletion in the typechecker

[bakkot]

Based on #612.

We don't (and are not going to) have generic types, so just hardcode these specific generic types. Also observe that a normal completion of never is equivalent to never, which allows us to get errors when passing NormalCompletion(x) to NormalCompletion(not-x). The latter doesn't really come up for arguments but it does for returned values, which I want to handle soon.

[michaelficarra]

We should add an assertNoTypeError test for a normal completion containing an ECMAScript language value and NormalCompletion(*true*) or something (the normal completion parameter type being a subtype of the declared type).

[bakkot]

We should add an assertNoTypeError test for a normal completion containing an ECMAScript language value and NormalCompletion(*true*) or something (the normal completion parameter type being a subtype of the declared type).

Done. Though our type system represents false as a distinct type from Boolean, so strictly speaking we already had that.

[michaelficarra]

I guess we could simplify that other test to a normal completion containing *false* if we wanted.