[Feature Request] Implement ZipLoRA Merge Algorithm

ZipLoRA, a novel method for merging low-rank adaptations (LoRAs) to achieve high-fidelity, concept-driven personalization in generative models, has shown promising results in maintaining style and subject integrity. I propose integrating this algorithm into sd-scripts to enhance its capabilities in generating personalized stylizations.

ZipLoRA leverages a "zipper-like" approach to blend two LoRAs, minimizing similar-direction sums and preserving original content and style properties. This is based on observations from recent generative model developments and the sparsity of LoRA weight matrices.

**Implementation References:**

- **Diffusers Implementation**: https://github.com/mkshing/ziplora-pytorch
- **Paper Page**: https://ziplora.github.io/

Start by reading the linked Diffusers implementation at github.com/mkshing/ziplora-pytorch and the ZipLoRA paper page to understand the algorithm and its expected inputs. Then locate the LoRA merging entry point in sd-scripts. Done means the algorithm is integrated into sd-scripts and can merge two LoRAs while preserving the stated style and subject properties.

Written by the indexing model from the issue text.