JavaScriptEventLoop: release the isSpinning latch with defer

`runAllJobs()` clears `queueState.isSpinning` only as its final statement, so
the flag survives as `true` if a job unwinds. `insertJobQueue` schedules a
drain only when `!isSpinning`, so after one unwound job the queue is never
drained again: every subsequent `enqueue` appends to a queue nothing will
run, for the lifetime of the process.

Nothing reports it. The failure is silent and total for asynchronous work,
while synchronous calls into the module keep working normally — which makes it
present as "async stopped" rather than as a crash.

Wrapping the reset in `defer` restores the invariant on every exit path. No
behaviour change on the normal path.

Author: mansbernhardt

Sources/JavaScriptEventLoop/JobQueue.swift

@@ -42,6 +42,15 @@ extension JavaScriptEventLoop {
 
     func runAllJobs() {
         assert(queueState.isSpinning)
+        // `defer`, so the latch is released even if a job throws.
+        //
+        // `runSynchronously` can unwind — a Swift runtime trap, or (under
+        // JavaScriptKit specifically) a JS exception crossing back into wasm.
+        // Clearing `isSpinning` only by falling off the end of the loop leaves
+        // it latched `true` on that path, and `insertJobQueue` then never
+        // schedules another drain: the executor is dead for the lifetime of the
+        // process, silently.
+        defer { queueState.isSpinning = false }
 
         while let job = self.claimNextFromQueue() {
             #if compiler(>=5.9)
@@ -50,8 +59,6 @@ extension JavaScriptEventLoop {
             job._runSynchronously(on: self.asUnownedSerialExecutor())
             #endif
         }
-
-        queueState.isSpinning = false
     }
 
     func claimNextFromQueue() -> UnownedJob? {