fix: make new codegen async realization-compatible

[Kha]

Follow-up to #7247

[leanprover-community-bot]

Mathlib CI status (docs):

• ❗ Batteries/Mathlib CI will not be attempted unless your PR branches off the nightly-with-mathlib branch. Try git rebase f8f1b2212ace2db4153e4b06db550931b4f093e3 --onto e3c6909ad593e5a966a449df5e92abb1f0dbc317. (2025-03-04 09:24:12)

[zwarich]

Putting this commit on top of master and then my new codegen branch causes a segfault during the stage2 build that is not obviously related to this change, but also nothing that I've seen before. I'll check that things still work as expected before the realizeConst changes and then try narrowing things down a bit more.

[zwarich]

Yeah, putting this on top of 0a55f4bf (which is before new calls to realizeConst, I think) leads to a segfault during the stage2 build with this backtrace:

* thread #3, stop reason = EXC_BAD_ACCESS (code=1, address=0x200081313c66)
  * frame #0: 0x00000001063d9810 libleanshared.dylib`l_Array_uget___boxed + 56
    frame #1: 0x000000010b63f6e4 libleanshared.dylib`lean::ir::interpreter::call(lean::name const&, lean::array_ref<lean::object_ref> const&) + 540
    frame #2: 0x000000010b63d834 libleanshared.dylib`lean::ir::interpreter::eval_body(lean::object_ref const&) + 3040
    frame #3: 0x000000010b63f9ec libleanshared.dylib`lean::ir::interpreter::call(lean::name const&, lean::array_ref<lean::object_ref> const&) + 1316
    frame #4: 0x000000010b63d834 libleanshared.dylib`lean::ir::interpreter::eval_body(lean::object_ref const&) + 3040
    frame #5: 0x000000010b641a90 libleanshared.dylib`lean::ir::interpreter::stub_m(lean_object**) + 636
    frame #6: 0x000000010b6391bc libleanshared.dylib`lean_object* lean::ir::interpreter::with_interpreter<lean_object*>(lean::elab_environment const&, lean::options const&, lean::name const&, std::__1::function<lean_object* (lean::ir::interpreter&)> const&) + 320
    frame #7: 0x000000010b641624 libleanshared.dylib`lean::ir::interpreter::stub_m_aux(lean_object**) + 92
    frame #8: 0x000000010b641180 libleanshared.dylib`lean::ir::interpreter::stub_6_aux(lean_object*, lean_object*, lean_object*, lean_object*, lean_object*, lean_object*) + 48
    frame #9: 0x000000010b656b48 libleanshared.dylib`lean_apply_2 + 896
    frame #10: 0x0000000109081cf8 libleanshared.dylib`l_Lean_Elab_liftMacroM___at_Lean_Elab_Tactic_evalTactic_expandEval___spec__1 + 1044

It seems that this is causing bad IR to be generated, written to an olean, and then executed later as part of a tactic.