test: make js worker queue unbounded

Author: joshua-spacetime

crates/core/src/host/v8/mod.rs

@@ -791,7 +791,7 @@ impl JsInstanceLane {
     /// If the worker disappears before replying, we replace it for future
     /// requests but surface the disconnect to the caller instead of retrying.
     /// This keeps instance-lane semantics closer to the old pooled-instance
-    /// model now that the worker queue is a rendezvous channel.
+    /// model while still preserving one-at-a-time execution on the worker.
     async fn run_once<R>(
         &self,
         label: &'static str,
@@ -1038,11 +1038,10 @@ async fn spawn_instance_worker(
     mut core_pinner: CorePinner,
     heap_policy: V8HeapPolicyConfig,
 ) -> anyhow::Result<(ModuleCommon, JsInstance)> {
-    // Spawn a rendezvous queue for requests to the worker.
-    // Multiple callers can wait to hand work to the worker, but with
-    // `bounded(0)` there is no buffered backlog inside the channel itself.
-    // The worker still processes requests strictly one at a time.
-    let (request_tx, request_rx) = flume::bounded(0);
+    // Spawn an unbounded queue for requests to the worker.
+    // Multiple callers can enqueue work without rendezvous handoff contention,
+    // but the worker still processes requests strictly one at a time.
+    let (request_tx, request_rx) = flume::unbounded();
 
     // This one-shot channel is used for initial startup error handling within the thread.
     let (result_tx, result_rx) = oneshot::channel();