Don't try to add an implicit undefined for mapped properties outside of strictNullChecks

[Andarist]

fixes #60390

[jakebailey]

I have a bad feeling that making isolated declaration emit more dependent on compiler options is a bad idea... Theoretically adding undefined here isn't a big deal because it "doesn't exist" when handled by the checker.

How does this end up fixing the assert?

[Andarist]

Hm, this isn't that relevant to ID (from what I understand). In the example, we want to generate types for an output of a factory function and ID requires explicit annotations - so it wouldn't even run into this problem.

The issue arises because you have a homomorphic mapped type (that preserves modifiers) running over prop?: undefined. So it ran into this code in requiresAddingImplicitUndefined:

                switch (declaration.kind) {
                    case SyntaxKind.PropertyDeclaration:
                    case SyntaxKind.PropertySignature:
                    case SyntaxKind.JSDocPropertyTag:
                        symbol ??= getSymbolOfDeclaration(declaration);
                        const type = getTypeOfSymbol(symbol);
                        return !!(symbol.flags & SymbolFlags.Property && symbol.flags & SymbolFlags.Optional && isOptionalDeclaration(declaration) && (symbol as MappedSymbol).links?.mappedType && containsNonMissingUndefinedType(type));

And then that led to obtaining an optional type for undefined (that's the type in the code above) but we never should try to do that with strictNullChecks: false. And in a sense, this proposed fix is solely based on that fact - we don't ever have to "add" undefined to anything when strictNullChecks: false.

[Andarist]

I could also fix this by avoiding adding undefined to... undefined but this seemed cleaner and it potentially skips some extra redundant work (although I perhaps could try to detect this in requiresAddingImplicitUndefined too)