[rmorehig]

Session and OAuth requests now discover the selected cluster through the existing /v1/clusters?organization=<name> endpoint and send its serving deployment headers automatically. Existing CLI commands, cluster paths, organization query parameters and the size catalog stay unchanged.

GET /v1/clusters?organization=acme
Authorization: Bearer <session>

{ "clusters": [{ "id": "...", "name": "production", "deployment_id": "12abcdef", "cloud_region": "us-east-1" }] }

POST /v1/query?organization=acme&cluster=production
Authorization: Bearer <session>
X-RawTree-Deployment-Id: 12abcdef
X-RawTree-Region: us-east-1

Routing metadata is cached for 30 seconds and invalidated by cluster/organization mutations. Both headers are omitted when metadata is absent or invalid, preserving the gateway migration fallback. API keys use embedded routing and skip discovery. Explicit request timeouts include discovery; gzip, bodies, idempotency keys and JWT/OAuth token contents retain their behavior.

Prepares CLI 0.7.0; publication is deferred. The organization-scoped endpoint migration is outside this PR.

Companions: Platform https://github.com/rawtreedb/rawtree-platform/pull/1524 and MCP rawtreedb/rawtree-mcp#34.

Validation: all CLI unit and contract tests pass, including routing, browser login and unchanged cluster URL assertions. Six live CLI tests pass against the rebuilt backend. P3 autoreview and thermonuclear maintainability review are clean.