Implemented support for Mistral Instruct 0.2 model.
Minor UX improvements in the chat app.
Multiple changes in various backend DLLs: better COM API to bind tensors for GPU (you need the updated version of the PackShaders.exe tool), image processor to decode bitmaps into tensors, improved support for importing PyTorch tensors (for example, can now compress FP32 tensors directly into BCML1)