- Introduced `chroma.hpp` with attention and modulation functions.
- Updated `CMakeLists.txt` to include Chroma tests.
- Enhanced `model.h` to support a new SD version for Chroma.
- Modified `stable-diffusion.cpp` to integrate T5 embeddings for Chroma, including padding mask generation.
- Adjusted condition retrieval logic to accommodate Chroma's unique requirements.

- Added step-by-step implementation plan for integrating the Chroma model into diffusion.cpp, focusing on layer-by-layer testing.
- Created CMakeLists.txt and Makefile for the new chroma_test example.
- Implemented main.cpp for testing VAE and T5 model loading and functionality via the Stable Diffusion API.
- Enhanced ModelLoader to support Chroma model weight loading and added debug logging for GGUF tensors.
- Updated stable-diffusion.cpp to integrate Chroma model handling, including condition handling and model-specific adjustments.
- Modified T5 and VAE classes to accommodate Chroma-specific configurations and parameters.

…lization

- Updated SD3CLIPEmbedder and FluxCLIPEmbedder to pass NULL for attention masks during T5 computation.
- Enhanced ChromaT5Embedder to generate and utilize a padding mask for T5 embeddings.
- Modified T5Stack and T5 classes to handle optional attention masks, allowing for NULL values.
- Adjusted T5Runner to make attention masks optional in the build_graph and compute methods.
- Added a new linear layer test example with comprehensive tensor operations and validation.
- Removed obsolete Makefile for chroma_test example.
- Updated main.cpp in chroma_test to reflect new model paths and configurations.

- Improved handling of attention_mask_01 to support various shapes and dimensions.
- Added checks for reshaping and permuting the attention mask based on input tensor dimensions.
- Enhanced logging for shape mismatches to aid debugging.
- Updated tensor reshaping logic to ensure compatibility with attention score broadcasting.
- Introduced model_dim as a member variable in the T5 class for better clarity and usage.
- Refined input tensor permutation logic to accommodate different embedding output shapes.

…ndling

…n.cpp into chroma