Add pile eval dataset

[rahul-tuli]

SUMMARY:
This PR adds the pile eval dataset to llmcompressor

TEST PLAN:
Added unit test for loading + ran the following script to verify

from llmcompressor.transformers import TextGenerationDataset, DataTrainingArguments
from transformers import AutoTokenizer

data_args = DataTrainingArguments(
            dataset="pile-eval",
            concatenate_data=True,
            splits={"calibration" : "validation[:1%]"},
        )
tokenizer = AutoTokenizer.from_pretrained("Xenova/llama2.c-stories15M")
dataset_manager = TextGenerationDataset.load_from_registry(
            data_args.dataset,
            data_args=data_args,
            split=data_args.splits,
            tokenizer=tokenizer,
        )


raw_dataset = dataset_manager.get_raw_dataset()
tokenized_dataset = dataset_manager.tokenize_and_process(
    raw_dataset, add_labels=True
)

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

[Satrat]

just had a few nitpicky comments, also why do we call it PileEval? from the source it looks like its actually "PileVal"

[kylesayrs]

LGTM