Add LLM routing fallback evidence gates

[minorstep]

/claim #1140

Summary

Updates llm-top-10 as an Improver-tier skill improvement for #1140. The change makes model routing, fallback behavior, context fitting, and hosted-model alias drift first-class evidence before LLM01-LLM10 scoring.

What changed

• Added a model routing and fallback evidence gate before LLM01-LLM10 scoring.
• Added route/fallback matrix fields for trigger, provider/runtime model ID, data class, tools, side-effect class, token/sampling policy, fallback target, fail-open/fail-closed mode, and change-control evidence.
• Added context-fitting checks so truncation/summarization cannot silently remove system/developer policy, authorization, tenant/source ACL metadata, denied actions, or active constraints.
• Added hosted-model alias/change-control checks for latest, preview, runtime model export, drift alerting, and regression tests before promotion.
• Expanded LLM01, LLM03, LLM06, and LLM10 guidance plus output tables and pitfalls for degraded routes.

Validation

• git diff --check
• Markdown fence-balance check
• Targeted content assertions for route/fallback, context-fitting, alias, fail-closed, and output-schema additions
• Private payout/payment-string scan of the changed skill file
• Live HTTP 200 checks for OWASP LLM Top 10 2025, OWASP LLM06 risk page, and OWASP LLM10 risk page
• Final pre-submit searches found no visible PR referencing [REVIEW] llm-top-10: add model router and fallback policy evidence #1140, model-router fallback policy, or LLM fallback policy.

Bounty note

This is a public Improver-tier submission. Payment details can be provided privately after maintainer acceptance.