PERRY_SHADOW_STACK=0 + PERRY_STATEPOINTS=1 silently emits a binary with no precise roots

[proggeramlug]

PERRY_SHADOW_STACK=0 combined with PERRY_STATEPOINTS=1 produces a binary with no precise frame roots at all, silently.

Measured: the statepoint + shadow-off build emits no __perry_gcmap section and is the same size as the plain shadow-off build — yet it ran and printed the correct answer, so nothing about the run suggests the roots are gone.

Why this happens, and why it matters more than the knob

reserve_shadow_slot() is the single entry point that, under native_stack_roots_enabled(), allocates a stack-map slot instead of a shadow-stack slot. Turning the shadow stack off therefore turns statepoints off with it.

Statepoints are an alternative lowering of the shadow stack's root-set analysis, not an independent mechanism. That has a consequence well beyond this knob: "delete the shadow stack and keep statepoints" is not currently expressible. Any plan that treats them as two interchangeable mechanisms — including the adoption plan in docs/engine-plan.md — rests on a premise the code does not support.

Asks

1. Make this combination a hard error. A rootless binary that runs and prints the right answer is the worst possible failure mode: it fails open, and nothing observable distinguishes it from a correct build until a collection frees a live object.
2. Record in the plan that the two mechanisms share their root-set analysis, so the "swap one for the other" framing gets corrected before anyone builds on it.

Found during the statepoint adoption soak.