v315: Let Scout identify the site and wait for its brief

[nicu-chiciuc]

Reviews started from a product name or several URLs could skip the startup brief, then continue while background research updated the site's metadata later. Remove URL extraction at startup: Scout identifies the main product with set_review_site, and both engines keep that call pending until the existing shared research finishes. The agent receives its private brief path or the actual research error before continuing.

Keep the six-minute research limit and show the existing preparation timer while the tool waits. Cached briefs, shared-job billing, Stop, and historical task snapshots retain their behavior. Reject a brief for the wrong hostname after an owner changes the review's site. No schema changes or additional queues.

Validation:

• pnpm run check: 1,665 tests passed; formatting, lint and all TypeScript checks passed.
• pnpm run build passed.
• Real Luna/Convex development review from “Open Checkvist and tell me what its free plan includes.” identified checkvist.com without a URL. Chrome showed the preparation timer and preserved it through refresh. Firecrawl completed after about six minutes; the tool returned the saved brief path, and Scout successfully read it with bash before continuing. No repeated model generations occurred during the wait.
• Separate code review covered both engines, cancellation, snapshot ownership, and deployment behavior.

Deployment: let all existing initial-start workflows finish before deploying, including Play tasks. Removing their old startup steps would otherwise cause a workflow journal mismatch. Do not merge while those tasks are running.