guardrails: codify MVP readiness now-vs-later split

[terisuke]

Parent epic: #1

Problem

The repo has a working guardrail MVP slice, but the remaining work is not yet classified into what blocks an MVP claim versus what can ship after the MVP floor is reached. That ambiguity makes issue #5 too broad and increases the risk of drifting away from the philosophy fixed in claude-code-skills epic #130, the source README, and the referenced skill-construction docs.

Deliverables

• local docs that split remaining MVP work into now required before MVP claim and later, after MVP floor
• missing issue briefs for current open implementation slices
• a GitHub epic and sub-issue structure that reflects the split
• updates to the top-level guardrails planning docs so future sessions start from the new structure instead of memory

Acceptance

• the repo has a written MVP readiness register
• GitHub has a dedicated epic for MVP readiness work
• guardrails: safe agents and workflow commands #5, guardrails: provider admission policy #6, guardrails: scenario and replay harness #7, and any newly needed work are explicitly grouped as now or later
• the canon from epic feat(guardrails): Wave 8 — review fixes + remaining hooks + multi-model delegation #130, the source README, and the Anthropic skill guide PDF remains the basis for implementation