Though I don't really understand why there is a distinction between the two self types. Do we use this captured_self_type field outside of the logic for invoke block? It seems like code inside the block would use the regular self_type.
It looks like in the PR we never upgrade the type with CapturedSelfOpnd, so it should always be Unknown

Hmm, yeah. I thought it could be useful when calling yield in two places, but in reality I didn't add any code for upgrading CapturedSelfOpnd. The initial reason to add this was to avoid a crash on StackOpnd(argc.try_into().unwrap()), so I could just directly fix it and leave captured->self type profiling not supported for now, assuming we often can't upgrade the type anyway.

addressed it in 38a6370