Temporarily disable the XLA sub-target for linear_operator_tridiag_test.

Due to the shard count increase, there are now some empty shards in the target,
since with XLA, some tests aren't dynamically added to the test classes,
which causes NO TESTS RAN failures on 3.12.

This only affects one test, as all others already require XLA to not be present.

PiperOrigin-RevId: 585715034

tensorflow/python/kernel_tests/linalg/BUILD

@@ -521,7 +521,9 @@ cuda_py_strict_test(
         "no_windows_gpu",
         "optonly",
     ],
-    xla_enable_strict_auto_jit = True,
+    # TODO(b/313470344): XLA temporarily disabled due to empty shards on 3.12.
+    xla_enable_strict_auto_jit = False,
+    xla_enabled = False,
     deps = [
         "//tensorflow/python/framework:config",
         "//tensorflow/python/framework:test_lib",