Generally we should avoid adding features to the interpreter that won't make the bootstrapping easier.
However implementing a compiler without type checking will be difficult, and we won't have a good debugging tools when things go wrong. So I think we need a type checker.
The type checker in the interpreter can be simpler than the type checker in the bootstrapped compiler. For example, we can probably get away with not having higher-kinded types.
We can also avoid polymorphic anonymous types (just have simple records and variants, or maybe only records).
Generally we should avoid adding features to the interpreter that won't make the bootstrapping easier.
However implementing a compiler without type checking will be difficult, and we won't have a good debugging tools when things go wrong. So I think we need a type checker.
The type checker in the interpreter can be simpler than the type checker in the bootstrapped compiler. For example, we can probably get away with not having higher-kinded types.
We can also avoid polymorphic anonymous types (just have simple records and variants, or maybe only records).