[Server] Separate UI and agent API endpoints

[Darkroom4364]

Current state:

• UI routes, operator API routes, WebSocket routes, and agent polling routes all live in the same server process and route namespace.
• This was acceptable for thesis/prototype testing, but it makes local operator controls and agent-facing endpoints harder to secure independently.

Why this matters:

• Operator-only APIs such as listener management, payload generation, file drop, logs, and server terminal need different trust assumptions than agent polling.
• Agent endpoints should stay minimal and stable, while operator endpoints can require authentication, CSRF/origin checks, and richer audit logging.

Acceptance criteria:

• Define an explicit route split for operator UI/API, WebSockets, and agent listener endpoints.
• Document which endpoints are intended to be exposed to agents and which are local/operator-only.
• Make the configured web/API port and listener ports unambiguous in docs and UI code.
• Add a small smoke test or documented manual test for command queueing after the split.

[Darkroom4364]

Completed via #93 and merged into dev. Operator UI/API/WebSocket routes now use an explicit operator mux, agent polling stays listener-owned, route ownership is documented, and the route-boundary checks passed in CI.