fix: reduce gpt-oss-120b GPU memory utilization to 0.90

Lower --gpu-memory-utilization from 0.95 to 0.90 to address CUDA OOM
crashes in vllm-gpt-oss containers under load.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Evrard-Nil

small-models.yaml

@@ -70,7 +70,7 @@ x-gpt-oss-common: &gpt-oss-common
   command: >
       openai/gpt-oss-120b
       --tensor-parallel-size 1
-      --gpu-memory-utilization 0.95
+      --gpu-memory-utilization 0.90
       --enable-prefix-caching
       --async-scheduling
       --max-num-seqs 64