test(tmproto): pin unknown-type and unknown-method wantErr substrings

[bokelley]

Closes #200.

Tightens two overly-broad wantErr substrings in tmproto/robustness_test.go that were too generic to catch wrong-error regressions:

• TestArtifactRef_Validate "unknown type": "type" → "not a known ArtifactRefType"
• TestAssetAccess_Validate "unknown method": "method" → "not a known AssetAccessMethod"

The old substrings matched any error containing the field-name prefix (artifact_ref.type, asset_access.method), so a regression that returned a different error class (e.g., "required" instead of "not a known …") would still pass. The new substrings pin the exact discriminating phrases already emitted by validate_ladder.go, making wrong-error regressions visible while keeping the no-caller-bytes invariant intact (wantNot guards are unchanged). No production code touched.

What tested

• go build ./... — clean
• go vet ./tmproto/... — clean
• go test ./tmproto/... ok 0.035s — all 195 tests pass including the two tightened cases
• go test ./... ok — full root-module suite passes

Pre-PR review

• code-reviewer: approved — new wantErr substrings confirmed as verbatim substrings of production errors; wantNot guards preserved; no production code changed; no blockers
• ad-tech-protocol-expert: approved — pins closed-enum rejection class correctly per TMP sanitization policy; no blockers

Triage-managed PR. This bot does not currently iterate on
review comments or PR conversation threads (only on the source
issue). To unblock:

• Push fixup commits directly: gh pr checkout <num> →
  fix → push.
• Or request a new first draft PR: comment /triage execute
  on the source issue only when no triage-managed PR is already
  open. Triage does not update existing PRs.

See adcp#3121
for context.

Session: https://claude.ai/code/session_01MjgZzyRVRrXju1ozznUSLT

---

Generated by Claude Code