feat: expand auditable policy experiments and organization autonomy

[Kristariss]

What changed

• add evidence-gated due diligence without exposing hidden quality labels
• add city imitation, enterprise exit, and government rescue dynamics
• add problem-first public LLM experiences with durable jobs and live progress
• replace exhaustive organization action whitelists with standard affordances, role capabilities, hard prohibitions, and bounded execution primitives
• document public experience boundaries and the organization-action security model

Why

InsideGov needs organization agents to generate novel strategies without allowing an LLM to invent authority, resources, evidence, or world outcomes. This PR keeps state settlement deterministic while expanding the policy-process action space.

Impact

Agents may propose new organization mechanisms. The authority engine classifies them as execute, execute with limits, requires coordination, requires approval, or blocked. Resource requests become approval tasks rather than direct world mutations.

Validation

• Backend: 99 tests passed
• Frontend: TypeScript check passed
• Frontend: production build and rendered HTML test passed
• Secret scan: no exposed DeepSeek API key patterns