For #4235 we introduced a workaround that forces the containing methods to jit when a block jits. However this is still not ideal, since if the containing method is only called once it will never see the jitted block. We would like to do a better job of jitting blocks in this situation, allowing them to run atop an interpreted containing method, but there are many issues here:

• Management of scopes and frames in the presence of JIT optimization.
• Structural differences in how interpreter versus JIT handles on-stack state

This would happen with post-9.2 performance work, ideally, so I'll target 9.2.1 for now.