Rust-based TrainingSample; accelerated, vectorised operations

[bghira]

The old VAE cache was doing a lot of Python threading gymnastics with queues and ThreadPoolExecutors. Replaced most of it with trainingsample's Rust batch operations since they're faster.

Changes:

• VAE cache now groups images by aspect ratio and processes them in actual batches
• Image resize/crop operations use ts.batch_resize_images() and ts.batch_*_crop_images()
• Added batch luminance calculation for metadata
• TrainingSample class uses trainingsample for resizing when possible
• Simplified concurrent image reading logic
• Kept all the fallbacks so it doesn't break if Rust explodes

Performance:

• ~6.8x speedup on image resize operations
• Less Python GIL contention
• Actually batches things instead of pretending to

Tested with SD3 PEFT LoRA training.

[Copilot]

Pull Request Overview

This PR replaces the Python-based VAE cache image processing with Rust-based batch operations from the new trainingsample library, delivering significant performance improvements. The implementation maintains backward compatibility through fallback mechanisms.

Key changes:

• Integrated trainingsample library for ~6.8x faster batch image operations
• Replaced complex Python threading with efficient batch processing grouped by aspect ratio
• Added new BatchedTrainingSamples class as interface to Rust operations

Reviewed Changes

Copilot reviewed 6 out of 9 changed files in this pull request and generated 4 comments.

File	Description
pyproject.toml, install/rocm/pyproject.toml, install/apple/pyproject.toml	Added trainingsample ^0.1.0 dependency across all platform configurations
helpers/image_manipulation/batched_training_samples.py	New wrapper class providing batch operations for resize, crop, and luminance calculations
helpers/image_manipulation/training_sample.py	Enhanced TrainingSample to use trainingsample for resize/crop operations with PIL fallbacks
helpers/caching/vae.py	Refactored VAE cache to use aspect-grouped batch processing instead of complex threading

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}