[booster] implement accelerator

[FrankLeeeee]

Overview

Accelerator is an abstraction for the hardware devices used for training. It can be CPU, GPU, TPU, etc. With this abstraction, we can provide room for future extension to more hardware which is compatible with PyTorch.

Wanna track the development progress? Take a look at

proposal: #3046
project kanban: API Refactoring

Goal

The Accelerator can move models to the target device. A sample definition is given below and subject to future changes.

class Accelerator:

        def __init__(self, device):
              self.device = device

        def setup_model(self, model) -> torch.nn.Module