v2.3.0

• Fix strength clip not properly normalized.
• Allow GRAY8 format for strength clip.
• Add dual parameter to perform inference in two threads for better performance. Mostly useful for TensorRT, not so useful for CUDA, and not supported for DirectML.