Process-backed schedulers need separate logical completion and physical drain

Reusable concurrency lesson from theft PR #814 / COR-4082 (observed 2026-09-08T22:23:59Z): in a process-backed decoder scheduler, cancellation can reject the logical result before the child process actually exits. If scheduler or request admission is released on result settlement, a winner can cause short-lived worker oversubscription while the losing process is still alive. Model execution with two promises: result for the caller-visible decision and a never-rejecting drained signal tied to the exact child generation's termination. Return the winning decode immediately, but retain per-root/global admission until every canceled sibling drains. Tests should use the real scheduler with controllable drain barriers; manually resolving a retained result can otherwise false-pass lane-release behavior. Source provenance: e3-solutions/theft PR #814, squash commit ae1d80cf3d4035184241d5132776f94909879847.