Queue::fake the last waitFor-racing parent/child summary test

Same bucket A pattern as the rest of the V2WorkflowTest waitFor cluster:
testWorkflowSummaryProjectsChildWaitAndHealthyRepairNoOp relied on the
testbench queue worker to drain the parent's start workflow task and the
child's start workflow task before the test could assert on the parent's
summary.wait_kind == 'child'. Under CI load this lost the race.

Converting to Queue::fake + drainReadyTasks lets the test deterministically
drive both runs through their first workflow task. drainReadyTasks already
respects available_at (d042ea9) so the child's 60s timer task is left
parked, leaving the parent in waiting_for_child as the test expects.

Clears the last non-skip MySQL CI failure under #399.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: durable-workflow-ops

tests/Feature/V2/V2WorkflowTest.php

@@ -2917,12 +2917,15 @@ public function testWorkflowCanWaitForChildWorkflowAndCompleteWithChildOutput():
 
     public function testWorkflowSummaryProjectsChildWaitAndHealthyRepairNoOp(): void
     {
+        Queue::fake();
+
         $workflow = WorkflowStub::make(TestParentWaitingOnChildWorkflow::class, 'parent-child-waiting');
         $workflow->start(60);
         $parentRunId = $workflow->runId();
 
         $this->assertNotNull($parentRunId);
 
+        $this->drainReadyTasks();
         $this->waitFor(static fn (): bool => $workflow->refresh()->summary()?->wait_kind === 'child');
 
         /** @var WorkflowLink $link */