fix(simulation): use non-blocking TCP to prevent game loop stall

read_framed() called read_exact() which blocked the entire bevy Update
schedule waiting for client input — no systems ran until a keystroke
arrived. Switch TcpStream to non-blocking mode so receive_inputs()
returns Ok(vec![]) on WouldBlock instead of blocking. Toggle to
blocking for snapshot writes (reliable delivery). Add 50ms frame
throttle (~20 ticks/sec) since the non-blocking loop would otherwise
spin. Downgrade input receive logging to trace, add error logging for
missing bridge resource and failed observer queries.

Fixes bug #1 (server never sends snapshots) and #2 (camera doesn't
center until first keystroke).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

server/src/bridge/mod.rs

@@ -73,7 +73,7 @@ pub fn receive_bridge_inputs(
     match bridge.receive_inputs() {
         Ok(inputs) => {
             for input in &inputs {
-                tracing::debug!(
+                tracing::trace!(
                     "Received input: tick={} action={:?}",
                     input.tick,
                     input.action
@@ -115,7 +115,10 @@ pub fn send_bridge_snapshot(
     mut buffer: ResMut<SnapshotBuffer>,
     mut running: ResMut<ServerRunning>,
 ) {
-    let Some(bridge) = bridge else { return };
+    let Some(bridge) = bridge else {
+        tracing::error!("send_bridge_snapshot: no BridgeResource");
+        return;
+    };
     if let Some(snapshot) = buffer.snapshot.take() {
         if let Err(e) = bridge.send_snapshot(&snapshot) {
             match &e {