Add thin agent context aggregation endpoint after read models stabilize

[cbusillo]

Problem

Skills should eventually call one compact Launchplane context surface instead of fanning out across work graph, products, Every Code, preview readiness, and deploy evidence, but that endpoint should not precede the lower-level contracts.

Scope

• Add an aggregation endpoint only after auth, redaction/provenance, repo mapping, work graph, Every Code summary, and preview readiness contracts are stable enough.
• Compose existing read models; do not create a parallel planning database.
• Key lookups by repo, branch, issue, PR, product, or work request when provided.
• Return compact context, source links, recommended next action, blockers, and unavailable/missing reasons.

Acceptance Criteria

• One read-only request can provide enough context for a skill preflight in common cases.
• The endpoint is a thin composition layer over tested lower-level read models.
• Responses degrade safely when parts of the context are unavailable or unauthorized.
• The contract is documented for public skill consumption.

Relationships

• Subissue of Make Launchplane a read-only agent context provider #379.
• Blocked by the lower-level read model and auth/redaction issues in this plan.