fix(proxy): stop serving routes whose owner has died

Problem

The daemon serves a route cache that only refreshes when routes.json changes, and loadRoutes drops dead owners only while loading. When an app's portless process is killed with SIGKILL it never touches the file, so its hostname keeps forwarding to the old port until some other route changes.

Changes

• The proxy checks the owner of the route a lookup resolves to, for HTTP requests, WebSocket upgrades and HTTP/2 extended CONNECT. A dead owner is skipped, so a wildcard request falls through to the next eligible parent, and the 404 page no longer lists it.
• Same policy as loadRoutes and prune, now one shared isRouteOwnerAlive: static aliases (pid 0) stay, and a process portless can't signal still counts as running (#498).
• Nothing is written to routes.json and no process is killed.
• Only the matched route is probed, so the cost is one process.kill(pid, 0) per request however many routes exist. Filtering the whole table per request cost about +80 µs at 50 routes and +565 µs at 200.

Closes #534. Complements #383, which prunes the file on a timer.

Visual proof

portless myapp node server.js, then kill -9 on the portless process (the dev server survives):

                        main    this branch
curl myapp.localhost    200     200
kill -9 <portless pid>
curl myapp.localhost    200     404
routes.json lists it    yes     yes

Written by an AI agent (Claude) on behalf of @jasonkuhrt.