docs(evidence): refresh EVIDENCE.md to v0.6.4 semantics + freshness tripwire

[ericckzhou]

@
Refreshes the artifact-contract doc where pre-0.6 wording went stale, and adds a coarse anti-entropy tripwire so the contract cant silently drift again. No runtime behavior changes.

Why

docs/EVIDENCE.md is the artifact contract — the document external readers trust about what a replay artifact is and what its verdicts claim. Stale claims there are higher-risk than a stale claim anywhere else in the repo. Several load-bearing places still carried pre-0.6 wording.

Changes — docs/EVIDENCE.md

• §4.5 — verdict-preservation list now carries all nine Verdict members (was a stale five), pointing to the §7 claim table.
• §2 — restated as the producer / read-only-consumer split (run / minimize produce; replay, inspect, diff, history, timeline, matrix, verify, export consume), replacing a mislabeled "five operations" list that actually had six bullets. doctor named as a diagnostic, not a protocol operation. Full command enumeration deliberately deferred to ARCHITECTURE.md to keep this doc protocol-level.
• §4.2 — materialized-spec lineage documents that paraphrase preserves validity_score / validity_method for each accepted rewrite.
• §4.1 — cli_invocation provenance cross-linked to its existing §6 semantic boundary rather than duplicated.
• §4.5 — clarifies semantic-oracle effects are preserved through the assigned verdict and existing case/judgment fields, not as a separate stored oracle-results payload.

New guard — tests/meta/test_evidence_doc_freshness.py

Coarse tripwire: every Verdict member and every read-only consumer must be named in the doc. The docstring is explicit that this is not a prose validator — a document-wide substring scan cannot catch intra-document inconsistency (e.g. a stale per-section list while §7 stays complete); that class stays a review concern.

Verification

• uv run pytest tests/meta/ → 17 passed
• uv run ruff check on the new test → clean
• "five operations" phrase confirmed removed
  @