Kernel tests for recipes are significantly slower than Functional tests

Based on my investigation into the performance regression, I've identified several contributing factors to the slowness in Recipe Kernel tests:

Both Functional and Kernel tests use MemoryCacheFactory as the default cache factory, but Kernel tests appear to be disproportionately impacted by its non-persistent nature. The frequent cache invalidations during module installation compound this issue.

Related to 📌 RecipeRunner::processInstall() installs modules one by one Active - every module installation triggers multiple cache invalidations. Even when replacing the cache factory with a persistent one in Kernel tests, there's no significant performance improvement due to this cache churn.

Attempted mitigation in our Kernel test:

public function register(ContainerBuilder $container) {
  parent::register($container);
  // Change container to database cache backends.
  $container
    ->register('cache_factory', 'Drupal\Core\Cache\CacheFactory')
    ->addArgument(new Reference('settings'))
    ->addMethodCall('setContainer', [new Reference('service_container')]);
}

XHProf profiling reveals that attribute discovery is the most time-consuming operation during Recipe Kernel tests:

• Majority of calls to getDefinitions() are pass-through calls from AttributeDiscoveryWithAnnotations::getDefinitions()
• 80% of these calls originate from DerivativeDiscoveryDecorator::getDefinitions()
• 80% of those calls come from DefaultPluginManager::findDefinitions()

For a more realistic comparison, I have removed idempotency check - so the second recipe application - from the GenericRecipeTestBase classes and captured the following with XHGui:

web/core/recipes/standard/tests/src/Kernel/GenericTest.php

web/core/recipes/standard/tests/src/Functional/GenericTest.php