fixed TensorType for TPU device execution, limited the unit test chec… (#3116)
* fixed TensorType for TPU device execution, limited the unit test check fro torch::kDouble on infinity entries
* typo fix
* improved the test
* integrated review comments
* linter fix
* patched the diff between master and this branch test/cpp/test_aten_xla_tensor.cpp. improved variable use in torch_xla/csrc/aten_xla_type.cpp
* linter fix