[torchlib] `native_batch_norm` needs more aten ops support

[titaiwangms]

1. func: _native_batch_norm_legit.no_stats(Tensor input, Tensor? weight, Tensor? bias, bool training, float momentum, float eps) -> (Tensor, Tensor, Tensor)
2. _native_batch_norm_legit_functional (generated by functionalization pass)
3. aten::to_copy_.default