allow cpu scalar to be moved to xpu in masked_fill (#103645)
# Motivation
Align to CUDA scenario, allow cpu scalar to be moved to xpu device in masked_fill.
# Solution
Add "xpu" support in condition control.
# Additional
no need for more UT.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103645
Approved by: https://github.com/jgong5, https://github.com/ezyang
