Add four unsigned fixture intake candidates for A2A #1576

[eriknewton]

This supplies the four fixture categories I offered in A2A #1576: dispute_dos, reputation_manipulation, cascade_refund, and skill_pricing_bait.

The fixtures are synthetic, unsigned intake candidates. Each specifies its policy, pre-state, attempted operation, expected refusal or review behavior, observable failure conditions, and a benign boundary control. They are deliberately separate from the admitted vectors/v0 corpus: I cannot produce AlgoVoi's signature, and the existing verifier, pinned key, signed vectors and manifest remain unchanged. The intake README documents maintainer signing and admission. The pending categories should remain pending until that admission is complete.

Validation covers fixture structure and policy consistency, benign controls, and rejection of unsigned candidates by the current verifier. These tests are not a live-rail conformance result. Existing canonical-corpus tests remain in place.

The intake does not replace the separate canonicalization, results-validation or negative-control proposals in #2, #3 and #4.

Validation: python -m pytest tests -q — 21 passed (14 existing, 7 new); python -m harness.validate_vectors --vectors vectors/v0 — all five existing signed vectors valid; git diff --cached --check — clean.