Support funasr/paraformer-zh to export OpenVINO IR model

[10vliu13]

I would like to support funasr/paraformer-zh to export OpenVINO IR model from optimum-intel. The original model implementation status is

1. The model network is implemented in python script file (https://github.com/modelscope/FunASR/tree/main/funasr) and covers SANMEncoder/ParaformerSANMDecoder/CifPredictor which separately in different script files in that repo.
2. The model.pt only contains model parameters in (https://huggingface.co/funasr/paraformer-zh)

As the original model in https://github.com/modelscope/FunASR/tree/main/funasr is complex, i reimplement the code into one python file naming modeling_paraformer.py. The implementation of modeling_paraformer.py is based on torch.nn.Module. Please see the below part of code for details. I went through the whole optimum-intel github and it seems there is no proper place to put this modeling_paraformer.py. @rkazants may I have your suggestion how to do that? Thanks

class Paraformer(torch.nn.Module):
"""
Author: Speech Lab of DAMO Academy, Alibaba Group
Paraformer: Fast and Accurate Parallel Transformer for Non-autoregressive End-to-End Speech Recognition
https://arxiv.org/abs/2206.08317
"""

def __init__(
    self,
    specaug: Optional[str] = None,
    specaug_conf: Optional[Dict] = None,
    normalize: str = None,
    normalize_conf: Optional[Dict] = None,
    encoder: str = None,
    encoder_conf: Optional[Dict] = None,
    decoder: str = None,
    decoder_conf: Optional[Dict] = None,
    ctc: str = None,
    ctc_conf: Optional[Dict] = None,
    predictor: str = None,
    predictor_conf: Optional[Dict] = None,
    ctc_weight: float = 0.5,
    input_size: int = 80,
    vocab_size: int = -1,
    ignore_id: int = -1,
    blank_id: int = 0,
    sos: int = 1,
    eos: int = 2,
    lsm_weight: float = 0.0,
    length_normalized_loss: bool = False,
    # report_cer: bool = True,
    # report_wer: bool = True,
    # sym_space: str = "<space>",
    # sym_blank: str = "<blank>",
    # extract_feats_in_collect_stats: bool = True,
    # predictor=None,
    predictor_weight: float = 0.0,
    predictor_bias: int = 0,
    sampling_ratio: float = 0.2,
    share_embedding: bool = False,
    # preencoder: Optional[AbsPreEncoder] = None,
    # postencoder: Optional[AbsPostEncoder] = None,
    use_1st_decoder_loss: bool = False,
    **kwargs,
):

    super().__init__()
    encoder = SANMEncoder(input_size=input_size, **encoder_conf)
    encoder_output_size = encoder.output_size()

    if decoder is not None:
        decoder = ParaformerSANMDecoder(
            vocab_size=vocab_size,
            encoder_output_size=encoder_output_size,
            **decoder_conf,
        )

    if predictor is not None:
        predictor = CifPredictorV2(**predictor_conf)

    self.encoder = encoder
    self.decoder = decoder
    self.predictor = predictor

def export(self, **kwargs):

    if "max_seq_len" not in kwargs:
        kwargs["max_seq_len"] = 512
    models = export_rebuild_model(model=self, **kwargs)
    return models