fix(poller): short-poll inbox instead of 30s long-poll

The 10s-after-healthy failure pattern was reproducing even with
the connection pool disabled. Smoking gun: the inbox loop opens
GET /messages?timeout=30 right after start_daemon returns, and
every subsequent peers/routes call timed out exactly when our
client-side reqwest timeout (10s) fired.

Concluded mycelium 0.6.1's HTTP server serialises requests: while
the long-poll connection is held, no other admin endpoint can
respond. The sidecar process kept logging routes the whole time
(seen in the in-app log buffer) — proof the daemon was alive,
just unable to serve concurrent calls.

Switch to short-poll: timeout=0 returns immediately, sleep 2s
between iterations. Per-iteration server hold time is now
millisecond-scale instead of 30s.

src-tauri/src/poller.rs

@@ -11,7 +11,7 @@ use tracing::warn;
 
 const PEERS_INTERVAL: Duration = Duration::from_secs(3);
 const ROUTES_INTERVAL: Duration = Duration::from_secs(5);
-const INBOX_LONG_POLL_SECS: u64 = 30;
+const INBOX_INTERVAL: Duration = Duration::from_secs(2);
 const INBOX_RETRY_BACKOFF: Duration = Duration::from_secs(2);
 const INBOX_CAPACITY: usize = 200;
 
@@ -121,18 +121,22 @@ fn spawn_inbox_loop(
 ) -> JoinHandle<()> {
     tokio::spawn(async move {
         loop {
+            tokio::time::sleep(INBOX_INTERVAL).await;
             let Some(client) = sidecar.client() else {
                 break;
             };
-            // Each iteration is a fresh long-poll. The daemon answers as
-            // soon as a message arrives, or returns an empty body / 204
-            // when the timeout window elapses.
-            match client.pop_message(false, INBOX_LONG_POLL_SECS, None).await {
+            // Short-poll: timeout=0 returns immediately if no message.
+            // We previously used a 30s long-poll, but mycelium 0.6.1's
+            // HTTP server appears to serialise requests behind a single
+            // worker — holding the connection for 30s starved every
+            // other endpoint (peers, routes, admin) until our own
+            // 10s reqwest timeout kicked in.
+            match client.pop_message(false, 0, None).await {
                 Ok(Some(msg)) => {
                     me.push_inbox(msg.clone());
                     let _ = app.emit("messages://incoming", &msg);
                 }
-                Ok(None) => {} // window expired, loop
+                Ok(None) => {}
                 Err(e) => {
                     warn!(error = %e, "inbox: pop_message failed");
                     tokio::time::sleep(INBOX_RETRY_BACKOFF).await;