[LTC] Pass a BackendDevice parameter into GetIrValueForScalarFromCodegen (#82970)
Summary: Currently GetIrValueForScalarFromCodegen uses CPU as the
default backend device for scalars, but we should make it a
backend-dependent decision.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82970
Approved by: https://github.com/Krovatkin, https://github.com/JackCaoG
