Adds check for EAI Harness differences in naming across versions

tasks.ALL_TASKS fails, it appears they changed it to tasks.all_tasks()
This adds flexibility for both ways.

src/backend/run_eval_suite_harness.py

@@ -34,7 +34,11 @@ def run_evaluation(eval_request: EvalRequest, task_names: list, num_fewshot: int
             "WARNING: --limit SHOULD ONLY BE USED FOR TESTING. REAL METRICS SHOULD NOT BE COMPUTED USING LIMIT."
         )
 
-    task_names = utils.pattern_match(task_names, tasks.ALL_TASKS)
+    try:
+        all_tasks = tasks.ALL_TASKS
+    except AttributeError:
+        all_tasks = tasks.all_tasks()
+    task_names = utils.pattern_match(task_names, all_tasks)
 
     logger.info(f"Selected Tasks: {task_names}")