This is a pretty effective test that we are stripping the user code out, but I'm not sure the cost benefit makes sense to land it. From what I can tell there is no real tangible benefit to building from the plan snapshot, but there are some real costs:

• persisting these data structures that we will have to support moving forward
• compat around changing the execution plan snapshot at all
  • In a world where we try to load from the plan snap, what happens if i execute from an a 0.10.0 dagit and a 0.10.X (where this is in) user process picks up execution? We likely pay some complexity cost to deal with this even if its swallowing all exceptions and fall back to recreating.

These costs are not enormous, but its enough to make me hesitant. I think its at least worth splitting this diff in to:

• all the good for user/host code changes

and

• @whitelist_for_serdes additions, plan snap changes, and the rebuild instead of re-create code path

We can discuss the recreate path more in its own diff and let all these other changes land right away.

putting in queue for now since host mode execution is not pressing from what I understand