Hosted relay on communities.buzz.xyz unreachable — Cloudflare 521, all relay HTTP endpoints time out

## Summary

A hosted community relay under `communities.buzz.xyz` has been unreachable for ~25 minutes. Cloudflare terminates TLS fine, but the origin relay never answers — the edge returns `521` on the WebSocket upgrade, and every relay HTTP endpoint times out.

Filing here for visibility since I couldn't find a status page. If this is the wrong venue for hosted-relay operational issues, happy to move it.

Relay host is masked below as `.communities.buzz.xyz`.

## Impact

The desktop app stays up, but every agent sits permanently disconnected — no messages in or out. Nothing is actionable from the client side.

## Timeline (UTC, 2026-07-28)

| Time | Event |
|---|---|
| ~00:06 | Existing agent sessions lose the relay, shut down cleanly |
| 00:08:14 | Agents restart, initial relay connect begins failing |
| 00:11:00 | Cloudflare starts returning explicit `521` on the upgrade |
| 00:16:16 | One transient `HTTP 200` on `/`, immediately back to timeouts |
| 00:30:39 | Still down — all endpoints time out |

## Evidence

### 1. Relay HTTP endpoints (from `crates/buzz-relay/src/router.rs`)

All time out with no response after a successful TLS handshake:

```
$ for p in health _liveness _readiness info; do
curl -s -o /dev/null -w "/$p %{http_code} %{time_total}s\n" \
"https://.communities.buzz.xyz/$p" --max-time 10
done
/health 000 10.003s
/_liveness 000 10.003s
/_readiness 000 10.004s
/info 000 10.002s
```

A WebSocket upgrade handshake (`Connection: Upgrade` / `Upgrade: websocket` / `Sec-WebSocket-Version: 13`) also returns `000`.

### 2. Cloudflare edge is healthy; the origin is not

```
$ nc -zvw 8 104.18.16.37 443
Connection to 104.18.16.37 port 443 [tcp/https] succeeded!
```

`curl -v` shows TCP connect + full TLS handshake completing, then no HTTP response until timeout. DNS resolves normally to `communities.buzz.xyz.cdn.cloudflare.net`.

Baseline from the same machine at the same moment:

| Host | Result |
|---|---|
| `.communities.buzz.xyz` | `000` (timeout) |
| `communities.buzz.xyz` | `000` (timeout) |
| `buzz.xyz` | `000` (timeout) |
| `block.xyz` | `200` |
| `cloudflare.com` | `200` |
| `github.com` | `200` |

So this is not local network or DNS — and it does not look scoped to a single community, since the apex and the `communities` host are down too.

### 3. `buzz-acp` client logs

```
00:08:44 WARN buzz_acp::relay: initial relay connect attempt 0 failed: Connection closed
00:09:15 WARN buzz_acp::relay: initial relay connect attempt 1 failed: Connection closed
00:11:00 WARN buzz_acp::relay: autonomous reconnect attempt 3 failed:
WebSocket error: HTTP error: 521
00:11:04 WARN buzz_acp::relay: autonomous reconnect attempt 4 failed:
WebSocket error: HTTP error: 521
```

Cloudflare `521` = "Web Server Is Down", i.e. the edge cannot reach the origin relay.

## Client behaviour is fine — no client bug here

Worth noting so this isn't misread as a client defect: `buzz-acp`'s reconnect logic behaves correctly throughout. After the inner 5-attempt burst it falls through to an outer reconnect loop with bounded, jittered backoff and keeps retrying indefinitely:

```
00:18:09 WARN buzz_acp::relay: relay reconnect failed: Connection closed
00:18:09 WARN buzz_acp::relay: retrying reconnect in 27.2s
00:19:06 WARN buzz_acp::relay: relay reconnect failed: Connection closed
00:19:06 WARN buzz_acp::relay: retrying reconnect in 52.3s
00:20:29 WARN buzz_acp::relay: relay reconnect failed: Connection closed
00:20:29 WARN buzz_acp::relay: retrying reconnect in 48.4s
```

Agents should reattach on their own once the origin is back. This issue is purely about the relay being down.

## Environment

- Desktop app `0.4.26`, macOS 15 (arm64)
- Relay: hosted `communities.buzz.xyz` deployment (not self-hosted)

## Ask

1. Is there a known outage / status page for hosted `communities.buzz.xyz` relays?
2. If a hosted relay is down, is there any in-app signal beyond agents silently showing as disconnected? Right now the only way to tell the difference between "relay down" and "my agents are broken" is reading `buzz-acp` logs.