Read the bounded, customer-scoped Agents aggregate

Returns one tenant-bound Agents DTO for SSR. It includes resolved static voice-stack display data, which components the customer actually configured (as opposed to the catalog defaults the resolved stack fills in), every language each agent answers in (selected_languages, which is the agent's own language alone for an agent whose owner never chose a language set -- language remains the one that OPENS the call), an inbound-admission verdict per agent, attached number routing, exact UTC-day activity, each agent's measured reply time (the median of its recorded worker-side reply turns, or an explicit unavailable when nothing was measured -- never a component-constant estimate presented as an observation), and wallet currency without provider or catalog API calls, verification, invoice, SMS, or webhook service reads. The admission verdict is a pure read of workspace state: admits/readiness reflect only conditions the inbound call path itself would refuse over; blockers is not a strict subset of those -- a small, explicitly declared set of codes (currently a misconfigured transfer_to_human destination) is real and actionable but does not affect admits/readiness, because invoking that tool mid-call fails gracefully rather than refusing the call. unevaluated lists the conditions the read could not evaluate before a call exists, and readiness is blocked or indeterminate — never ready, because the gates that need the agent's number are not run here. A line-scoped readiness is reported by the Overview aggregate's primary_line. An agent answered by the customer's own service is judged on the endpoints it needs rather than on the managed stack it does not use. Soft-deleted agents are not part of the workspace: they are excluded from agents and from total_agents.