Add sd lora data container and preprocessing funcs

[xiaoyu-work]

This pull request introduces a new sd_lora data component to the Olive data pipeline, adding advanced image preprocessing and auto-captioning capabilities for Stable Diffusion LoRA training. The main changes include registering the new component, implementing aspect ratio bucketing for efficient batching, and supporting automatic image captioning using BLIP-2 and Florence-2 models.

New SD LoRA Data Component:

• Registered the new sd_lora module in olive.data.component.__init__.py, making it available for import and use in the Olive data pipeline.
• Added a copyright header to the sd_lora package to ensure proper licensing.

Image Preprocessing Enhancements:

• Implemented aspect_ratio_bucketing in sd_lora/aspect_ratio_bucketing.py, which automatically assigns images to buckets based on aspect ratio and resolution, supporting resizing and cropping for efficient Stable Diffusion training. Includes utilities for bucket generation, image resizing, and crop coordinate calculation.

Auto-Captioning Support:

• Added auto_caption, blip2_caption, and florence2_caption preprocessing functions in sd_lora/auto_caption.py, enabling automatic image captioning using BLIP-2 and Florence-2 models. Supports batch processing, device selection, overwrite logic, and flexible caption storage (file or in-memory).

These additions significantly enhance Olive's data preparation workflow for image generation and LoRA fine-tuning tasks.## Describe your changes

Checklist before requesting a review

• Add unit tests for this change.
• Make sure all tests can pass.
• Update documents if necessary.
• Lint and apply fixes to your code by running lintrunner -a
• Is this a user-facing change? If yes, give a description of this change to be included in the release notes.

(Optional) Issue link

[github-advanced-security]

lintrunner found more than 20 potential problems in the proposed changes. Check the Files changed tab for more details.

The base branch was changed.