debug: check full HD for chunk1 test

tests/unit/test_fmha_6warp_tma_multirow_multitile.cu

@@ -180,7 +180,7 @@ static int test_single(int T, int s_k, int n_h = 1, int batch = 1) {
             o_ref, nullptr, HD, T, s_k, SCALE);
 
         float cs = 0, na = 0, nb = 0;
-        int check_hd = HD_CHUNK;  // Only check first hd_chunk values (for partial debug)
+        int check_hd = HD;  // Check full HD (all chunks)
         for (int t = 0; t < T; t++) {
             for (int d = 0; d < check_hd; d++) {
                 float a = bf16_to_f32_host(h_o[h * MAX_T * HD + t * HD + d]);