fix: the relay check reads the watchdog's probe; 0.4.9.12:5 → 0.4.9.12:6

[stupleb]

Tor logs a notice for every control connection it accepts. The Relay Reachability check from #33 opened its own every 30 seconds beside the watchdog's. A relay's log went from 120 to 248 New control connection opened lines an hour. On the relay where I measured it, those lines are 88% of a 10,000-line log. That log now covers under two days, down from about three and a half.

Change

• probe() carries the relay check's three GETINFO keys in the round trip it already makes.
• relayStatus() returns that reading. It returns null when the latest probe went unanswered or is older than 90 seconds.
• The relay check still runs every 30 seconds, so it still fails at once when the Public address is turned off.
• README: names the SocksPort warning Tor logs on every start.

Verified. I ran the real control.ts, before and after this change, against Tor 0.4.9.12 over a unix control socket, offline:

• five watchdog ticks plus five relay reads: 10 control connections before, 5 after, with identical readings;
• a 551 Address unknown reply mid-batch leaves the later replies intact;
• null before any probe, null after a probe Tor didn't answer, null past 90 seconds;
• tsc, prettier and the x86 build pass.

Verified on a StartOS box (x86_64):

• identical 240-second windows with the relay on gave 16 control connections on 0.4.9.12:4 and 8 with this change;
• the check reports success while reachable, fails at once when the Public address is turned off, and succeeds again when it is turned back on.

That ran before rebasing onto #34, which does not touch control.ts or relay.ts.

Considered and rejected: Log [control]warn [~control]notice stdout would remove the remaining 120 lines an hour. It also hides Tor's Bootstrapped N% lines, which are logged in the same domain.

[helix-a]

The shared probe preserves the relay health-check semantics while eliminating its duplicate Tor control connection. The stale-reading guard also handles missed watchdog probes cleanly. Verified locally with npm ci, npm run check, Prettier, npm run build, and make x86; all passed.

[helix-a]

GitHub’s required-signature rule blocked the merge despite the passing checks and approval. I re-signed the existing commit without changing its tree (tree e32c5fc1fdc8e79eb0e03d479fd7e6ee2394faac), preserving Stuart as author, then re-ran this approval against the signed head. The code remains exactly what I reviewed.