[TransformersPipeline] Add in and refactor TransformersPipeline args

[dsikka]

For this ticket: https://app.asana.com/0/1201735099598270/1205276886236966/f

Summary:

• Updates the TransformersPipeline constructor to add in two arguments: config and tokenizer
• For both of these, the user can provide a string, path or transformers object which will be used as opposed to relying on a deployment directory with the expected json files. By default, these will both be None and in that case, the normal deployment directory workflow will be used.
• Additionally, the config argument may also be a dictionary
• To support this functionality, the get_onnx_path_and_configs is refactored/separated into two separate functions, get_hugging_face_configs and get_onnx_path

Testing:

• Tested locally using a variety of combinations for config and tokenizer

Example:

from deepsparse import Pipeline
from transformers import LlamaConfig, LlamaTokenizerFast

tokenizer = LlamaTokenizerFast.from_pretrained("hf-internal-testing/llama-tokenizer")
config = {
   "_name_or_path": None,
   "architectures": [
      "LlamaForCausalLM"
   ],
   "bos_token_id": 1,
   "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 5120,
   "initializer_range": 0.02,
   "intermediate_size": 13824,
   "max_position_embeddings": 4096,
   "model_type": "llama",
   "num_attention_heads": 40,
   "num_hidden_layers": 40,
   "num_key_value_heads": 40,
   "pretraining_tp": 1,
   "rms_norm_eps": 1e-05,
   "rope_scaling": None,
   "tie_word_embeddings": "false",
   "torch_dtype": "float16",
   "transformers_version": "4.31.0.dev0",
   "use_cache": "true",
   "vocab_size": 32000
}

llama = Pipeline.create(
   task="text-generation",
   model_path="/home/dsikka/models_llama/deployment_13",
   engine_type="onnxruntime",
   deterministic=False,
   config=config,
   tokenizer=tokenizer
)

inference = llama(sequences=["Hello?"])
for s in inference.sequences:
   print(s)