[NPU] be compatible with auto-round & initial auto-round support

[rnwang04]

Description

1. Why the change?

https://github.com/analytics-zoo/nano/issues/1786

2. User API changes

If we have an auto-round quantized model (Llama / Qwen / Minicpm / Baichuan) whose bits is 4 & group size is -1, we can load it and run it on NPU with below usage :

from ipex_llm.transformers.npu_models.convert_auto_round_model import convert_auto_round_model_to_npu_model
from transformers import AutoModelForCausalLM, AutoTokenizer
from auto_round import AutoRoundConfig
import torch

if __name__ == "__main__":
    backend = "cpu"
    quantization_config = AutoRoundConfig(
        backend=backend,
    )

    model = AutoModelForCausalLM.from_pretrained(quantized_model_path,
                                                 device_map=backend.split(':')[0],
                                                 quantization_config=quantization_config,
                                                 attn_implementation="eager",
                                                 trust_remote_code=True,
                                                 torch_dtype=torch.float16)
    print("finish load quantized model.")

    model = convert_auto_round_model_to_npu_model(model, save_directory=save_dir)

    tokenizer = AutoTokenizer.from_pretrained(original_model_path, trust_remote_code=True)

Note, to run auto-round models, below is the env requirement:

conda create -n npu-round python-3.11
conda activate npu-round
# for NPU
pip install --pre --upgrade ipex-llm[npu]
# for auto-round
pip install auto-round==0.4.2
pip uninstall auto_gtpq

3. Summary of the change

• Support auto-round-patch to load auto-round model on cpu for Windows
• Support unpack_auto_round_layer for sym_int4 & asym_int4
• Support necessary code changes to be compatible with auto-round
• Support converting auto-round model to NPU optimized model in convert_auto_round_model_to_npu_model function

This is just an initial PR, we can enhance it or integrate it into from_pretrained later if necessary.

4. How to test?

• Unit test: Please manually trigger the PR Validation here by inputting the PR number (e.g., 1234). And paste your action link here once it has been successfully finished.