Previously, boxed tuples with different type identifiers wouldn't be
considered as having the same identity even when containing the same
objects, for example when comparing `(U32(0), U32(0))` through
`(U32, U32)` and `(Any, Any)` tuples.

The new identity algorithm for boxed values follows. "unchanged" means
that this step of the algorithm didn't need any changes.

- Check if the addresses are the same (unchanged)
  - If they are, the identities are the same (unchanged)
  - Otherwise, check the kind of the LHS object
    - If it's an unboxed type, the identities are different (unchanged)
    - If it's a boxed number, check the type of the RHS object
      - If the types are the same, check the values via `memcmp`
        (unchanged)
      - Otherwise, the identities are different
    - If it's a boxed tuple, check the kind of the RHS object
      - If it's a boxed tuple of any type, check the values via the
        LHS's `__is` function
      - Otherwise, the identities are different

The algorithm of the `__is` function was also changed. Since it now
needs to compare against any possible type, multiple cases are
generated in the function, which each case unboxing to a particular
type and doing the comparison. The case to execute is chosen based on
the type ID of the RHS object.

Fixes ponylang#1950.