Generalize meta tensor pipeline

[mrwyattii]

DeepSpeed-Inference has supported meta tensor loading with BLOOM, OPT, GPT-J, and GPT-NEOX models for some time, but this functionality was never extended to MII. Adding that support here by refactoring the BloomPipeline to work with other models.

Motiviated by #198

Example of added meta_tensor option in mii_config:

import mii
mii_configs = {"tensor_parallel": 8, "dtype": "fp16", "meta_tensor": True}
mii.deploy(task="text-generation", model="facebook/opt-66b", deployment_name="opt", mii_config=mii_configs)