[Entrypoint] Update oneshot to post_train

[horheynm]

Blocked on #1109 to obtain models without training_args

SUMMARY:

• Update oneshot to post_train
• Make post_train have its own pipeline, decoupled form main
• Make post_train's input arguments only dependent on ModelArguments, DatasetArguments, and RecipeArguments. Any TrainingArguments-related input arg will raise an error. Ex. overwrite_output_dir.
• Update any oneshot usage in the function name (run_oneshot_and_finetune), Enum attributes (RecipeStages), variable name (oneshot_model) to post_train`.
• Update folder names from oneshot to post_train (ex. "tests/llmcompressor/transformers/oneshot/oneshot_configs")
• Update cli command in setup.py
• In stage-runner, in the logic to detect if post_train or train, change the order to detect post_train first (train is a subset, so currently train will always be detected if post_train).

TEST PLAN:

• Pass existing tests
• search oneshot using grep.

FOLLOWUPS:

• Remove oneshot dependent code (ex. SessionMixin.apply)
• Remove stage-runner

[github-actions]

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

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.

[horheynm]

flip the order check if the config_name has "post_train" first before "train".

[horheynm]

Only used in training args, not post-train dep args. Will raise an error if present

[dsikka]

we should be landing the entrypoint updates, independent of changing the naming from oneshot to post_train.

[kylesayrs]

Can you please redirect this to post_train with a deprecation warning?