Added tuned kernel block size for customer's model (#1071)

Signed-off-by: cychiuak <andersonchiu@google.com>

Author: cychiuak

tpu_inference/kernels/ragged_paged_attention/v3/tuned_block_sizes.py

@@ -1231,6 +1231,13 @@
                 },
             }
         },
+        16: {
+            'q_bfloat16_kv_bfloat16': {
+                'q_head-8_kv_head-1_head-128': {
+                    262144: (128, 256),
+                }
+            }
+        },
     },
     'TPU v5e': {
         128: {