Post-lock hygiene from the round-3 referee's non-blocking notes

The round-3 referee ratified the gate-1 lock (PR #33, comment 4885159605) with non-blocking notes, deliberately left out of the lock to avoid re-refereeing a converged pre-registration. None changes a locked threshold; all are hygiene on the machinery around it. Land after the lock merges:

1. Derivations test: assert mean/sd consistency with the values arrays. tests/test_gates_derivations.py binds thresholds to artifact mean/sd fields, but a hand-edited mean/sd with an intact values[] would pass unnoticed. Add an assertion that each floor stat's mean/sd equals the recomputed statistics of its values array (ddof=1), for every artifact cited by a derivations.floor_run.
2. Remove the stale round-1-era motivation clause at gates.yaml lines ~43-49 (pre-amendment text describing the runs view as the chained-model catch; superseded by the persistence-attribution note in the gate-1 block).
3. Declare computed-but-ungated quantities. The runs view's energy/tail stats, the change-moment mean, and transition_rates are computed by the harness but neither locked nor listed in reported_not_gated. List them there explicitly so nothing is silently ungated.

Context: rounds 1-3 of the pre-lock review are comments 4884803805, 4885102386, 4885159605 on PR #33.

Start with tests/test_gates_derivations.py and the referenced derivations.floor_run artifacts; run the existing derivation tests before changing them. Then inspect gates.yaml around lines 43-49 and the gate-1 block. Done means consistency assertions cover every cited artifact, stale motivation text is removed, and the listed computed quantities appear in reported_not_gated.