What does this PR do?

Fixes #14091

Two resource cleanup bugs in the serve/web commands and the SSE event endpoint:

1. serve.ts and web.ts — server.stop() is dead code (unreachable after await new Promise(() => {})), and no signal handlers exist. On SIGTERM/SIGINT, Instance.disposeAll() never runs, so MCP subprocesses, LSP servers, and other child resources are orphaned.

2. server.ts — The SSE /event endpoint puts clearInterval(heartbeat) and unsub() inside stream.onAbort(), which Hono only calls on client disconnect. When the server closes the stream on InstanceDisposed via stream.close(), cleanup never runs and the interval + subscription leak.

Changes

serve.ts and web.ts — Add signal handlers following the existing pattern from worker.ts:137-147:

const shutdown = async () => {
  await Promise.race([Instance.disposeAll(), new Promise((resolve) => setTimeout(resolve, 5000))])
  server.stop(true)
  process.exit(0)
}
process.on("SIGTERM", shutdown)
process.on("SIGINT", shutdown)

server.ts — Add clearInterval(heartbeat) and unsub() to the InstanceDisposed handler before stream.close(). Both calls are idempotent, so the existing onAbort handler remains as a fallback for normal client disconnects.

How did you verify your code works?

• Full test suite passes (996 tests, 0 failures)
• Manual: bun dev serve, then kill -TERM <pid> — process now exits cleanly (exit 0) instead of being killed by signal (exit 143)
• Code review: the shutdown pattern is identical to the existing worker.ts implementation that already handles this correctly