Fix FLA fused recurrent generation cache state

fla/ops/gsa/fused_recurrent.py

@@ -101,12 +101,13 @@ def fused_recurrent_gsa_inference(
     else:
         hk0, hv0 = q.new_zeros(B, H, K, M, dtype=torch.float), q.new_zeros(B, H, M, V, dtype=torch.float)
 
-    hkt, hvt = None, None
-    if output_final_state:
-        if NG == 1:
-            hkt, hvt = hk0, hv0
-        else:
-            hkt, hvt = q.new_empty(B, H, K, M, dtype=torch.float), q.new_empty(B, H, M, V, dtype=torch.float)
+    # The Triton inference kernel stores final recurrent state for the first
+    # query head in each KV group. It therefore needs valid hkt/hvt pointers
+    # even when the caller does not request returning the final state.
+    if NG == 1:
+        hkt, hvt = hk0, hv0
+    else:
+        hkt, hvt = q.new_empty(B, H, K, M, dtype=torch.float), q.new_empty(B, H, M, V, dtype=torch.float)
 
     o = v.new_empty(B, T, HQ, V)
     grid = (B * HQ,)