Don't block IceAdapter.start() on telemetry websocket connect (fixes #42)

TelemetryDebugger.startupComplete() called websocketClient.connectBlocking()
synchronously on the IceAdapter startup thread, with no timeout. Telemetry
is a debug-only observability channel - its data flow is strictly outgoing
(see ice-adapter/src/main/java/com/faforever/iceadapter/telemetry/, all
messages implement OutgoingMessageV1) and no game logic depends on it being
connected - so it should never gate peer connectivity.

When the telemetry server is silently unreachable (TCP layer up but app
layer hung - for example an alive load balancer fronting a dead backend),
the WebSocket Upgrade handshake never receives an HTTP 101 response.
connectBlocking() then waits until TCP keepalive eventually kills the
socket, about two hours later. During that whole window IceAdapter.start()
is blocked, the FAF client's adapter-ready orchestration trips its own
timeouts, and the user-visible symptom is "the ice adapter doesn't connect
to other players".

Fix: run the connect on a virtual thread so startup returns immediately.
The pre-existing reconnect loop in sendingLoop() already handles transient
drops, and queued messages catch up once the socket opens because the
messageQueue is unbounded.

Also adds the missing return after the InterruptedException catch
(previously fell through to sendMessage on an interrupted thread) and
re-asserts the interrupt flag, per Java best practice.

Fixes #42

Author: juerghegglin

ice-adapter/src/main/java/com/faforever/iceadapter/debug/TelemetryDebugger.java

@@ -115,18 +115,29 @@ private void sendingLoop() {
 
     @Override
     public void startupComplete() {
-        try {
-            if (!websocketClient.connectBlocking()) {
+        // Connect to the telemetry server asynchronously: telemetry is a debug-only channel and
+        // must NOT block IceAdapter.start(). Previously connectBlocking() ran on the startup
+        // thread with no timeout, so when the telemetry server's TCP layer was reachable but
+        // its application layer was hung, the WebSocket Upgrade response never arrived and
+        // startup waited indefinitely (until TCP keepalive — ~2 hours). The peer-connectivity
+        // path was held the whole time, which is why the adapter "didn't connect" when
+        // telemetry was down. See https://github.com/FAForever/java-ice-adapter/issues/42
+        Thread.ofVirtual().name("telemetry-connect").start(() -> {
+            try {
+                if (!websocketClient.connectBlocking()) {
+                    Debug.remove(this);
+                    return;
+                }
+            } catch (InterruptedException e) {
                 Debug.remove(this);
+                log.error("Failed to connect to telemetry websocket", e);
+                Thread.currentThread().interrupt();
                 return;
             }
-        } catch (InterruptedException e) {
-            Debug.remove(this);
-            log.error("Failed to connect to telemetry websocket", e);
-        }
 
-        sendMessage(new RegisterAsPeer(
-                UUID.randomUUID(), "java-ice-adapter/" + IceAdapter.getVersion(), IceAdapter.getLogin()));
+            sendMessage(new RegisterAsPeer(
+                    UUID.randomUUID(), "java-ice-adapter/" + IceAdapter.getVersion(), IceAdapter.getLogin()));
+        });
     }
 
     @Override