PM Resize LoRA

Resizes a LoRA to a different rank using tensor decomposition. This node reduces or increases the rank of all layers in a LoRA model using SVD-based methods. Lower ranks reduce memory usage and may reduce strength, while maintaining semantic meaning. Decomposition Methods: - SVD: Full singular value decomposition (slow but optimal) - rSVD: Randomized SVD (fast, recommended for most cases) - energy_rSVD: Energy-based randomized SVD (best for DiT/large LoRAs) The resizing uses asymmetric singular value distribution (all S values in up matrix) which differs from the symmetric distribution used in lora_decompose.