refactor: Use llama_memory_state_ptr for child states in hybrid memory state

Branch: HybridRecurrentCache

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

Author: gabe-l-hart

src/llama-kv-cache-hybrid-recurrent.cpp

@@ -244,9 +244,9 @@ const llama_ubatch & llama_kv_cache_hybrid_recurrent_state::get_ubatch() const {
 }
 
 const llama_kv_cache_unified_state * llama_kv_cache_hybrid_recurrent_state::get_state_attn() const {
-    return state_attn.get();
+    return static_cast<const llama_kv_cache_unified_state *>(state_attn.get());
 }
 
 const llama_kv_cache_recurrent_state * llama_kv_cache_hybrid_recurrent_state::get_state_recurrent() const {
-    return state_recurrent.get();
+    return static_cast<const llama_kv_cache_recurrent_state *>(state_recurrent.get());
 }

src/llama-kv-cache-hybrid-recurrent.h

@@ -145,6 +145,6 @@ class llama_kv_cache_hybrid_recurrent_state : public llama_memory_state_i {
     std::vector<uint32_t>     heads_attn;
     std::vector<llama_ubatch> ubatches;
 
-    const llama_kv_cache_unified_state_ptr   state_attn;
-    const llama_kv_cache_recurrent_state_ptr state_recurrent;
+    const llama_memory_state_ptr state_attn;
+    const llama_memory_state_ptr state_recurrent;
 };