Remove the Cursor worker from devbox-server (and the spurious yellow 'degraded' status it causes)

[louisreingold]

Why

We've moved devbox-server to Claude sessions as the way to drive work in an environment. The Cursor self-hosted worker is now dead weight:

• It autostarts per environment but mostly just sits there / disconnects.
• When its process isn't alive, status.js marks the whole environment degraded (the yellow dot) — even though WordPress and Claude sessions work perfectly. That's confusing: a fully usable box shows yellow because of an unused worker.

So: rip out the Cursor worker functionality from the server, and with it the worker-liveness component of env status (no more yellow-because-worker-dead).

What to remove (server only)

• server/src/worker.js — delete (Cursor worker start / liveness / health).
• server/src/fleet.js — delete (Cursor fleet-API enrichment; only exists for the worker).
• server/src/manager.js — drop worker autostart in the create pipeline, the worker start in start(), worker supervision in the reconcile loop, and the worker-health gathering in _gather().
• server/src/status.js — remove the worker-liveness branch. After this, core-up ⇒ running. (Keep degraded only if we still want it for partial core services up — see open questions.)
• server/src/config.js — remove cursorWorkerAutostart, workerDir, workerManagementAddr, workerIdleReleaseTimeout, fleetApiUrl. Decide on cursorApiKey (see below).
• server/src/routes.js / status publicView — drop the worker and fleet fields from env responses.
• server/ui/app.js — remove any worker/fleet display.
• Docs: server/README.md, server/.env.example, server/AGENT_USAGE.md — strip Cursor-worker config + mentions.

Open questions for the implementer

1. CURSOR_API_KEY: it's currently a required server env var, but the worker is its only consumer. Once the worker is gone, make it optional / remove it from the required-vars check.
2. degraded status: keep it to mean "some but not all core services up" (genuine partial failure), or collapse to just running/stopped/failed? Leaning: keep degraded for the partial-core case, just remove the worker cause.

Out of scope

The scaffolder's Cursor support is separate and stays: cursor-agent baked into the workspace image and the npm run cursor launcher (a human can still run Cursor interactively in a devbox). This issue is only about the server's worker orchestration.

Not starting on this now — filing for later.