Perf: allow whole-run nursery backing reclamation

Downstream: zig-utils/zig-js#427 Follows: #17, #20, #32, #33

Disposition: rejected and reverted

The generic prototype grouped consecutive dead same-size cells only for authoritative owned iterators, withdrew classification for the complete run before finalizers, and released storage afterward. Generic bindings retained the per-cell path. It passed the full normal and suppression-free TSan suites as 64d2660.

The required downstream order-balanced exact-parent A/B then rejected the design: eight-lane sweep improved only 0.7%, wall time regressed 0.9%, the candidate won 2/7 pairs, and one lane regressed 1.1%. The hook was unused after zig-js reverted its activation, so zig-gc removed the speculative API in 88ea254, restoring the exact tested parent source.

Completed experiment

• Define and test the quiescent owned dead-run contract and generic fallback.
• Preserve finalization, weak processing, payload indexing, stable IDs, accounting, full/minor collection, and teardown ordering.
• Pass full normal and suppression-free TSan dependency suites.
• Run the required downstream 1/2/4/8 exact-parent A/B with exact checksums.
• Revert completely when the material-improvement gate failed.
• Publish the downstream raw evidence and verdict.

Downstream implementation and exact A/B are now tracked by zig-utils/zig-js#427.

Closed as a measured dead end. The generic contract was correct, but the downstream performance gate failed, so 88ea254 removes the unused API and restores the tested parent tree.