Parent: #336 Related: #1, #144, #359, #360
Close the remaining deterministic no-GIL overlap evidence gap without widening the supported movement boundary. A running or parked peer must make both direct and scheduled moving attempts fail closed; the exact same pending request may move only after every peer exits.
Scope:
- create a true no-GIL precise-GC context and warm a real baseline numeric tier;
- park a spawned Thread on the existing Atomics publication/wait protocol;
- from a native callback, protect a survivor, prove direct compaction is unsupported, and schedule movement;
- while the peer remains parked, run the warmed native tier through multiple movement-safe checkpoints and prove the request remains pending, no relocation token opens, and no movement counter advances;
- after signaling and joining the peer, consume the same request quiescently and prove the protected payload moves with intact semantics;
- machine-check the running-thread gate and document the supported/fail-closed matrix.
Acceptance:
- deterministic running and parked peer overlap cannot observe forwarding or stale cells;
- the request is neither lost nor prematurely cleared while peers exist;
- post-join direct compaction reports useful movement and a dense second pass;
- Debug, ReleaseSafe, suppression-free TSan, C/C++/Objective-C hosts, and relocation/release inventories are green.