[Misc] Refactor get_kv_cache_spec into AttentionLayerBase (#26587)

Signed-off-by: NickLucche <nlucches@redhat.com>

vllm/model_executor/models/deepseek_v2.py

@@ -481,7 +481,7 @@ class DeepseekV32IndexerCache(torch.nn.Module, AttentionLayerBase):
             raise ValueError(f"Duplicate layer name: {prefix}")
         compilation_config.static_forward_context[prefix] = self
 
-    def get_kv_cache_spec(self) -> KVCacheSpec:
+    def get_kv_cache_spec(self, vllm_config: VllmConfig) -> KVCacheSpec:
         return MLAAttentionSpec(  # Only has one vector instead of K + V
             block_size=self.cache_config.block_size,
             num_kv_heads=1,