[DTensor] Fix device detection logic for TestDTensorPlacementTypes::test_split_tensor. (#105357)
The test should respect self.device_type as it checks whether the environment
has enough GPUs to serve the requested world size.
The test will lead to hangs if we try to run 8 ranks over our 2-4 GPUs CI instances.
Fixes #104769
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105357
Approved by: https://github.com/wanchaol
