Launch a reconnaissance job for a pentest.

The launch is attributed to the dedicated recon machine user, not to the caller — the requester's identity is recorded in the audit trail, not on the launch row.

201 means the recon has been scheduled, not that artiacker-manager has accepted it: the row is claimed here and POSTed by the dispatch task the claim triggers, so the response carries a PENDING launch whose status is the thing to read afterwards. A cluster that later declines it is recorded as ERROR on that launch, not returned here.

Every failure maps through the centralized handlers, so nothing is caught here: an unknown or out-of-scope slug is 404, a Cerbos denial 403, no wildcard in scope 400, an already-running recon 409, and an unreachable artiacker-manager — for the already-running check, the one call still made inline — 500.

Args: pentest_slug (str): Public pentest identifier. db (AsyncSession): Request-scoped database session. context (ActiveContext): Resolved caller identity and scope.

Returns: SuccessResponse: The legacy create envelope carrying the launch id.