Stage2 misc fixes

[Vexu]

I'm not 100% sure about the second commit but it does make sense based on my understanding of the current intended semantics of packed structs. And for the generic poison check, the second best alternative I could come up with was to add the check to resolveInst which would have been a much more intrusive change.

Closes #11709
Closes #11680
Closes #11550
Closes #11504

[andrewrk]

the second best alternative I could come up with was to add the check to resolveInst which would have been a much more intrusive change.

Hm, why would that be more intrusive? There is typically only a handful of calls to resolveInst at the top of each ZIR instruction handler, and only one per operand. On the other hand, typeOf is needed in utility functions that operate on Air.Inst.Ref parameters, and often typeOf is called multiple times per Air.Inst.Ref. Furthermore, most of the backends already have a fallible resolveInst including the LLVM backend and the x86_64 backend.

[Vexu]

Hm, why would that be more intrusive? There is typically only a handful of calls to resolveInst at the top of each ZIR instruction handler, and only one per operand. On the other hand, typeOf is needed in utility functions that operate on Air.Inst.Ref parameters, and often typeOf is called multiple times per Air.Inst.Ref. Furthermore, most of the backends already have a fallible resolveInst including the LLVM backend and the x86_64 backend.

Hmm, I might have estimated wrong, I'll give it a try.