Typed Racket pretends mutable sets don't exist, and yet it assumes all hashtables could be mutable

Typed Racket only provides types for immutable set operations. It would be painless to add types for mutable operations as well, but TR also considers Setof types to be covariant, which obviously isn't true with mutable sets.

This is inconsistent with how TR handles hashtables. In an ideal world, I'd like HashTable to be immutable and have a separate MutableHashTable for mutable hash tables and mutable operations. This would be nice for a number of reasons:

• This would let hashtables be covariant.
• This would make contracts on immutable hashtables flat contracts, so predicate generation and occurrence typing would work with them.
• Mutability would need to be explicitly noted in the type system.

I'd unfortunately guess that this is not possible to change at this point. Still, the behavior with sets and hashtables are inconsistent. If we can't change HashTable, can we make MutableHashTable and ImmutableHashTable subtypes? Then could we make the equivalent variations for sets?

The issue names no files, tests, or entry points. Start by locating Typed Racket's set and hashtable type definitions and their variance handling. Done would require an agreed design for mutable and immutable collection types, plus corresponding operations and tests.