fix: ship interim Sol 6.1 routing and catalog freshness guards

[Brian Krabach (bkrabach)]

Summary

Replace all 39 OpenAI-family Terra candidates and nine Copilot Terra pins with exact Sol 6.1, preserving ordering/efforts. Keep current clean GPT-6 Luna for utility/UI roles; no pre-v6 curated fallback. Refresh Copilot Claude/vision pins, preserve caller model/effort/origin mount and instance_id, add bounded catalog checks including Gemini GA-Pro successors. This is an approved opportunistic update; proper comparative quality/cost evaluations follow separately.

Verification

• Complete offline source suite in an isolated source-installed Python 3.13 environment with native Rust Core: 729 passed. No added test-file exclusions. Existing platform/live skips are reported separately.
• Combined ecosystem check: 9,422 passed, six existing skips, zero failures.
• Catalog-backed replay: 8 matrices, 219 candidates, 64 provider topologies, 6,656 direct role resolutions; zero missing exact pins.
• Fresh four-provider catalog check: 211 candidates, zero findings.
• Eight real Core hook mounts and 468 default-derived ChatGPT child plans: zero strict violations.
• Small Sol 6.1 API compatibility trials at low/medium/high/xhigh/max plus ChatGPT high passed. These are request smokes, not quality benchmarks.
• Fresh-context semantic/privacy review completed.

Boundaries

No credential provisioning, authenticated live-workflow dispatch, enable-variable changes or user configuration updates. Scheduled live catalog CI stays disabled pending dedicated credentials. Explicit user model pins remain respected.

Generated with Amplifier

Co-Authored-By: Amplifier 240397093+microsoft-amplifier@users.noreply.github.com

Intentional regression data update

Regenerated the existing golden fixture for approved model-policy changes; all eight recorded matrices, guard assertions and exemption sets are retained. Legacy Terra stays in caller classification, not stock role candidates.