This PR fixes an issue when running Mathlib's `FintypeCat` as code,
where an erased type former is passed to a polymorphic function. We were
lowering the arrow type to`object`, which conflicts with the runtime
representation of an erased value as a tagged scalar.
This PR changes `IRType.boxed` to map `erased` to `tobject` rather than
`object`, since `erased` has a representation of a boxed scalar 0 when
we are forced to represent it at runtime. This case does not occur at
all in the Lean codebase.
This PR changes ToIR to call `lowerEnumToScalarType?` with
`ConstructorVal.induct` rather than the name of the constructor itself.
This was an oversight in some refactoring of code in the new compiler
before landing it. It should not affect runtime of compiled code (due to
the extra tagging/untagging being optimized by LLVM), but it does make
IR for the interpreter slightly more efficient.
This PR enables code generation to proceed in parallel to further
elaboration.
It does not aim to make further refinements such as generating code for
different declarations in parallel or removing the dependency on kernel
checking.
