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

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

python/sglang/srt/models/gpt_bigcode.py

@@ -87,6 +87,7 @@ class GPTBigCodeAttention(nn.Module):
             scaling=self.scale,
             num_kv_heads=self.num_kv_heads,
             layer_id=layer_id,
+            quant_config=quant_config,
             prefix=add_prefix("attn", prefix),
         )