fix(metrics): recognize approval_requested/replanned as honest-red outcome tokens

[SollanSystems]

Closes #36.

Adds approval_requested and replanned to _HONEST_RED_OUTCOME_TOKENS in scripts/metrics.py — they are known, non-success outcomes emitted by the sanctioned writer (loop/emit.py _ITERATION_OUTCOMES), so a RUNLOG written entirely through emit.append_iteration no longer surfaces them under provenance.unrecognized_outcomes. "replan" is retained for prose-token variants (it never matched "replanned" — different string).

Round-trip regression test added to scripts/test_metrics.py: iterates the live emit._ITERATION_OUTCOMES tuple (not a hardcoded copy, so future vocabulary drift re-triggers it), appends one iteration per token, asserts provenance.unrecognized_outcomes == []. The test unlinks the scaffolded RUNLOG first so it does not depend on the #40 fix landing.

Verification (independent adversarial pass):

• Negative control: the new test overlaid on an unfixed main archive fails with AssertionError: ['approval_requested', 'replanned'] == [] — the test bites.
• The honest-red set feeds only the provenance surface: FCR derives from the unchanged completion/progress token sets, RP from repair records — zero FCR/RP numbers change anywhere (docs/metrics-baseline.json untouched).
• Gates: targeted pair 62 passed; full suite 394 passed / 14 skipped (engine-extra env guards run in CI); self_eval 13/13.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.