Update Legacy OpenLLM leaderboard to use "train" split for ARC fewshot (#2802)

* Update openllm.yaml to use train fewshot split for arc

lm_eval/tasks/benchmarks/README.md

+### Changelog
+- 2025-Mar-17 OpenLLM v2: Fixed few-shot split to correctly use train set for arc_challenge.

lm_eval/tasks/benchmarks/openllm.yaml

@@ -2,7 +2,7 @@ group: openllm
 group_alias: Open LLM Leaderboard
 task:
   - task: arc_challenge
-    fewshot_split: validation
+    fewshot_split: train
     num_fewshot: 25
   - task: hellaswag
     fewshot_split: train
@@ -16,3 +16,5 @@ task:
     num_fewshot: 5
   - task: gsm8k
     num_fewshot: 5
+metadata:
+  version: 2