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

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

python/sglang/srt/models/minicpm3.py

@@ -192,6 +192,7 @@ class MiniCPM3Attention(nn.Module):
             self.scaling,
             num_kv_heads=self.num_local_heads,
             layer_id=layer_id,
+            quant_config=quant_config,
             prefix=add_prefix("attn", prefix),
         )
 
@@ -343,6 +344,7 @@ class MiniCPM3AttentionMLA(nn.Module):
             num_kv_heads=1,
             layer_id=layer_id,
             v_head_dim=self.kv_lora_rank,
+            quant_config=quant_config,
             prefix=add_prefix("attn", prefix),
         )