Fix loading KV quantization scale; Enable modelopt kv cache (#4686)

Co-authored-by: qingquansong <ustcsqq@gmail.com>

python/sglang/srt/models/exaone.py

@@ -155,6 +155,7 @@ class ExaoneAttention(nn.Module):
             self.scaling,
             num_kv_heads=self.num_kv_heads,
             layer_id=layer_id,
+            quant_config=quant_config,
         )
 
     def forward(