Some helper code for determining missing meta coverage for XLA ops
When I ran it I got this:
```
$ PYTORCH_COMPARE_XLA=/scratch/ezyang/xla/xla_native_functions.yaml python test/test_meta.py
aten.logdet.default
aten._local_scalar_dense.default
aten.cholesky.default
aten.diag.default
aten.empty.memory_format # SKIP
aten.index.Tensor
aten.kthvalue.default
aten.masked_select.default
aten.max_pool3d_with_indices.default
aten.max_unpool2d.default
aten.max_unpool3d.default
aten.native_batch_norm.default # SKIP
aten.nll_loss2d_forward.default
aten.nonzero.default
aten.prelu.default
aten.relu.default
aten.roll.default
aten.rrelu_with_noise.default
aten.std_mean.correction
aten.symeig.default
aten.take.default
aten.trace.default
aten.native_layer_norm.default # SKIP
```
Signed-off-by: Edward Z. Yang <ezyangfb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/78464
Approved by: https://github.com/albanD
