llama : allocate indexer cache only in "full" indexer layers (#26474)

Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>

src/llama-kv-cache-dsa.h

@@ -26,7 +26,8 @@ public:
                      uint32_t   n_pad,
                      uint32_t   n_swa,
                llama_swa_type   swa_type,
-        const layer_filter_cb & filter,
+        const layer_filter_cb & filter_mla,
+        const layer_filter_cb & filter_lid,
         const  layer_reuse_cb & reuse);
 
     ~llama_kv_cache_dsa() = default;