[Chore] Bump lm-eval version (#31264)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

.buildkite/lm-eval-harness/run-lm-eval-chartqa-vllm-vlm-baseline.sh

@@ -2,7 +2,7 @@
 # We can use this script to compute baseline accuracy on chartqa for vllm.
 #
 # Make sure you have lm-eval-harness installed:
-#   pip install lm-eval==0.4.9
+#   pip install "lm-eval[api]>=0.4.9.2"
 
 usage() {
     echo``