[gemini] gemini supports lazy init

[ver217]

📌 Checklist before creating the PR

• I have created an issue for this PR for traceability
• The title follows the standard format: [doc/gemini/tensor/...]: A concise description
• I have added relevant tags if possible for us to better distinguish different PRs

🚨 Issue number

Link this PR to your issue with words like fixed to automatically close the linked issue upon merge

e.g. fixed #1234, closed #1234, resolved #1234

Closes #3529

📝 What does this PR do?

Summarize your work here.
if you have any plots/diagrams/screenshots/tables, please attach them here.

• Fix NVME optimizer
• Refactor Gemini and make it support ColoInitContext, LazyInitContext and naive initialize.
• Update Gemini plugin.

As env of CI is not compatible with MetaTensor, lazy init test is skipped. I tested on local machine:

💥 Checklist before requesting a review

• I have linked my PR to an issue (instruction)
• My issue clearly describes the problem/feature/proposal, with diagrams/charts/table/code if possible
• I have performed a self-review of my code
• I have added thorough tests.
• I have added docstrings for all the functions/methods I implemented

⭐️ Do you enjoy contributing to Colossal-AI?

• 🌝 Yes, I do.
• 🌚 No, I don't.

Tell us more if you don't enjoy contributing to Colossal-AI.

[kurisusnowdeng]

There is an issue caused by importing colossalai

Python 3.9.16 (main, Mar  8 2023, 14:00:05) 
[GCC 11.2.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import colossalai
Traceback (most recent call last):
  File "/home/.conda/envs/dev/lib/python3.9/site-packages/torch/_ops.py", line 565, in __getattr__
    op, overload_names = torch._C._jit_get_operation(qualified_op_name)
RuntimeError: No such operator aten::prelu_backward

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/ColossalAI/colossalai/__init__.py", line 1, in <module>
    from .initialize import (
  File "/home/ColossalAI/colossalai/initialize.py", line 18, in <module>
    from colossalai.amp import AMP_TYPE, convert_to_amp
  File "/home/ColossalAI/colossalai/amp/__init__.py", line 11, in <module>
    from .apex_amp import convert_to_apex_amp
  File "/home/ColossalAI/colossalai/amp/apex_amp/__init__.py", line 4, in <module>
    from .apex_amp import ApexAMPOptimizer
  File "/home/ColossalAI/colossalai/amp/apex_amp/apex_amp.py", line 13, in <module>
    from colossalai.nn.optimizer import ColossalaiOptimizer
  File "/home/ColossalAI/colossalai/nn/__init__.py", line 1, in <module>
    from ._ops import *
  File "/home/ColossalAI/colossalai/nn/_ops/__init__.py", line 1, in <module>
    from .addmm import colo_addmm
  File "/home/ColossalAI/colossalai/nn/_ops/addmm.py", line 6, in <module>
    from ._utils import GeneralTensor, Number, convert_to_colo_tensor, reduce_grad, reduce_input
  File "/home/ColossalAI/colossalai/nn/_ops/_utils.py", line 7, in <module>
    from colossalai.nn.layer.utils import divide
  File "/home/ColossalAI/colossalai/nn/layer/__init__.py", line 7, in <module>
    from .moe import *
  File "/home/ColossalAI/colossalai/nn/layer/moe/__init__.py", line 1, in <module>
    from .checkpoint import load_moe_model, save_moe_model
  File "/home/ColossalAI/colossalai/nn/layer/moe/checkpoint.py", line 5, in <module>
    from .experts import MoeExperts
  File "/home/ColossalAI/colossalai/nn/layer/moe/experts.py", line 12, in <module>
    from colossalai.zero.legacy.init_ctx import no_shard_zero_decrator
  File "/home/ColossalAI/colossalai/zero/__init__.py", line 1, in <module>
    from .gemini import (
  File "/home/ColossalAI/colossalai/zero/gemini/__init__.py", line 3, in <module>
    from .gemini_ddp import GeminiDDP, ZeroDDP
  File "/home/ColossalAI/colossalai/zero/gemini/gemini_ddp.py", line 17, in <module>
    from colossalai.utils.model.experimental import LazyTensor
  File "/home/ColossalAI/colossalai/utils/model/experimental.py", line 10, in <module>
    from colossalai._analyzer._subclasses import MetaTensor
  File "/home/ColossalAI/colossalai/_analyzer/_subclasses/__init__.py", line 1, in <module>
    from ._meta_registration import *
  File "/home/ColossalAI/colossalai/_analyzer/_subclasses/_meta_registration.py", line 277, in <module>
    aten.prelu_backward.default,
  File "/home/.conda/envs/dev/lib/python3.9/site-packages/torch/_ops.py", line 569, in __getattr__
    raise AttributeError(
AttributeError: '_OpNamespace' 'aten' object has no attribute 'prelu_backward'

I'm using Pytorch 2.0 & CUDA 11.8