Add audit provenance for agent-initiated context and intents

[cbusillo]

Problem

Agent-triggered reads and write intents need enough audit/provenance to explain who requested what, why Launchplane allowed it, which source facts were used, and what external mutation or GitHub action occurred.

Scope

• Define audit fields for agent subject, request source, action, resource, safety level, authorization decision, idempotency key, and resulting record links.
• Include provenance in context and intent responses without exposing secrets.
• Make denied or unavailable actions visible enough for operators to diagnose safely.

Acceptance Criteria

• Agent-initiated write intents produce durable audit/evidence links.
• Read/context responses include enough provenance to distinguish Launchplane records, GitHub facts, provider evidence, and worker reports.
• Denials include safe reason codes.
• Tests cover audit records or response provenance for representative allowed and denied actions.

Relationships

• Subissue of Make Launchplane a read-only agent context provider #379.
• Supports scoped write intents and remote-user access.