[MTIA Aten Backend] Migrate _log_softmax.out / _log_softmax_backward_data.out (pytorch#156539)

# Context

See the first PR pytorch#153670

# This diff

 Migrate _log_softmax.out / _log_softmax_backward_data.out to in-tree.

Differential Revision: [D77044380](https://our.internmc.facebook.com/intern/diff/D77044380/)

Pull Request resolved: pytorch#156539
Approved by: https://github.com/malfet
ghstack dependencies: pytorch#156502

Author: andyanwang

aten/src/ATen/native/native_functions.yaml

@@ -3720,6 +3720,7 @@
   dispatch:
     CPU: log_softmax_cpu_out
     CUDA: log_softmax_cuda_out
+    MTIA: log_softmax_mtia_out
     MPS: log_softmax_mps_out
 
 - func: _log_softmax_backward_data(Tensor grad_output, Tensor output, int dim, ScalarType input_dtype) -> Tensor
@@ -3730,6 +3731,7 @@
   dispatch:
     CPU: log_softmax_backward_cpu_out
     CUDA: log_softmax_backward_cuda_out
+    MTIA: log_softmax_backward_mtia_out
     MPS: log_softmax_backward_mps_out
 
 - func: _logcumsumexp(Tensor self, int dim) -> Tensor