feat: alora migration to peft

[YashasviChaurasia]

Description of the change

This PR updates the codebase to switch from the external alora package to the native Activated LoRA (aLoRA) support that is now part of the Hugging Face peft library.
Key changes:

• Removing legacy alora imports and wrappers.
• Updating config_utils.py to handle aLoRA fields via peft.LoraConfig.
• Adjusting tests to use alora_invocation_tokens for validation and inference.
• Simplifying run_inference.py by removing custom aLoRA handling (now handled natively by PEFT).

Related issue number

How to verify the PR

Was the PR tested

• I have added >=1 unit test(s) for every new method I have added.
• I have ensured all unit tests pass

[github-actions]

Thanks for making a pull request! 😃
One of the maintainers will review and advise on the next steps.

[dushyantbehl]

Minor changes.

[YashasviChaurasia]

@dushyantbehl Currently moving the --alora_invocation_string to LoraConfig is causing issue for the following reasons:

if peft_config is not None and hasattr(peft_config, "alora_invocation_string"):
    inv_str = getattr(peft_config, "alora_invocation_string")
    if not inv_str:
        raise ValueError(
            "`--alora_invocation_string` is required when using --peft_method alora."
        )
    ...

As now we use LoraConfig for both alora and lora runs setting up alora_invocation_string as a dataclass attribute in LoraConfig triggers the above code snippet causing it to fail as it is set to None by default if not provided by the user..
It seems simpler to keep it outside Dataclass and use it via additional as to avoid conflicts with Lora testcases.

[ashokponkumar]

Good work @YashasviChaurasia !

[dushyantbehl]

LGTM