metal : another fix for the fa kernel

ggml/src/ggml-metal.metal

@@ -2144,6 +2144,7 @@ kernel void kernel_flash_attn_ext_f16(
                     const short tx = tiisg%4;
                     const short ty = tiisg/4;
 
+                    if (iq1 + ty < ne01) {
                         // mqk = mqk*scale
                         ss[8*cc + ty*TF + 2*tx + 0] *= scale;
                         ss[8*cc + ty*TF + 2*tx + 1] *= scale;
@@ -2160,6 +2161,7 @@ kernel void kernel_flash_attn_ext_f16(
                         }
                     }
                 }
+            }
 
             // used to detect blocks full of -INF
             float smax = -INFINITY;