[BE][DTensor] merge random init test to test_random_ops.py (#98874)
Random Ops behavior becomes different on CUDA device and others in PR https://github.com/pytorch/pytorch/pull/98199 . As a result, test of DTensor random initialization in test_init.py was skipped. This PR fixes it by having different testing assertion for different types of device and also merges existing duplicated test.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98874
Approved by: https://github.com/wanchaol