[CI][Bugfix] Fix Hybrid SSM NixlConnector PD prefix cache test (2 GPUs) (#47157)

Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>

tests/v1/kv_connector/nixl_integration/run_mamba_prefix_cache_test.sh

@@ -51,6 +51,7 @@ vllm serve $MODEL \
   --trust-remote-code \
   --enable-prefix-caching \
   --mamba-cache-mode all \
+  --attention-backend FLASHINFER \
   --kv-transfer-config "$KV_CONFIG" &
 
 # Start decode instance
@@ -68,6 +69,7 @@ vllm serve $MODEL \
   --trust-remote-code \
   --enable-prefix-caching \
   --mamba-cache-mode all \
+  --attention-backend FLASHINFER \
   --kv-transfer-config "$KV_CONFIG" &
 
 echo "Waiting for prefill instance on port $PREFILL_PORT..."