[Chat pipeline] session context manager

[bfineran]

updated implementation of #1239

includes small bug fix for text gen input parsing due to behavior with pydantic aliases

requested UX:

with chat_pipeline.session():
    first_result = chat_pipeline(prompt=PROMPT, *args, **kwargs)
    second_result = chat_pipeline(prompt=PROMPT, *args, **kwargs)

test_plan:
unit test included and ran locally, but disabled due to overhead in GHA

manual verification as well:

from deepsparse import Pipeline

pipeline = Pipeline.create(
    "chat",
    model_path="/home/benjamin/neuralmagic/models/codegen_mono-350m-bigpython_bigquery_thepile-pruned50/deployment",
    engine_type="onnxruntime"
)

with pipeline.session():
    output_1 = pipeline(prompt="first", generation_config=dict(max_new_tokens=1))
    output_2 = pipeline(prompt="second", generation_config=dict(max_new_tokens=1))

output_3 = pipeline(prompt="third", generation_config=dict(max_new_tokens=1))

output_2.session_ids == output_1.session_ids  # expected True
output_3.session_ids == output_1.session_ids  # expected False

Co-authored-by: @rahul-tuli