top-k.cu

This release introduces DFlash2 support with local convolution and candidate selection, alongside various bug fixes and code refactoring. Key changes include optimizations for DFlash2 and fixes for mrope and graph number calculations.

This release focuses on CUDA optimization by replacing the init_offsets kernel in argsort with iterators, alongside minor cleanup in CUDA source files. It also provides extensive pre-compiled binaries for multiple platforms.