download compressed deployment directory

[bfineran]

to land with neuralmagic/sparsezoo#373

updates deployment directory downloads to target the compressed directory instead of loose files

simple test:

from deepsparse import Pipeline

pipeline = Pipeline.create(
    "text-generation",
    model_path="zoo:nlg/text_generation/mpt-7b/pytorch/huggingface/mpt_chat/base-none",
    engine_type="onnxruntime"
)

pipeline("test", generation_config=dict(max_new_tokens=10))