test(claims): CI coverage for claim 3 rug-pull + claim 4 nonce binding

[imran-siddique]

What

Adds the pytest coverage for claim 3 (rug-pull detection) and claim 4 (TRACE Claim nonce binding) that the README CI table and the paper's evaluation section flagged as pending.

Closes #353, closes #354.

Tests

tests/unit/test_claim3_rug_pull_detection.py (6):

• definition + catalog hash determinism
• avalanche (>64/256 bits) on a one-sentence description tamper
• aggregate catalog hash changes on a single-tool tamper
• CatalogHashMismatch raised fail-closed when a tampered catalog is loaded under the approved pinned hash
• approved catalog passes its own pinned hash
• tamper is undetectable without pinning (the malicious sentence loads; only the pin turns it into a block)

tests/unit/test_claim4_trace_claim_nonce.py (6):

• nonce determinism
• session binding (nonce changes with session_id)
• instance binding (nonce changes with TEE key)
• a session-A claim's nonce does not match session-B's expected nonce
• session_id tamper breaks the Ed25519 signature
• audit-entry removal breaks the export signature

Patterned after the existing test_claim1_hash_binding.py / test_claim2_session_gap.py suites and the claim 3/4 experiment scripts.

Verification

In a clean venv against main: ruff check src/ tests/ clean, pytest on claims 3+4 → 12 passed.

🤖 Generated with Claude Code