libmir-cuda 0.3.0

CUDA inference backend for libmir
mod dense;
mod gated;
mod kernel;
mod nvfp4;
mod pair;
mod reduce;
#[cfg(all(test, target_os = "linux"))]
mod tests;

pub use dense::{
    DenseExpertCanonicalizer, DenseGateUpLayout, DenseGatedActivation, SelectedDenseDispatch,
    SelectedDenseGateLaunch, SelectedDenseMoe, SelectedDenseMoeSpec, SelectedDenseReduceLaunch,
};
pub use gated::{
    GatedActivation, SelectedAffineGated, SelectedAffineGatedLaunch, SelectedAffineGatedSpec,
};
pub use nvfp4::{NvFp4BankView, SelectedNvFp4Gated, SelectedNvFp4Reduce, SelectedNvFp4Spec};
pub use pair::{SelectedAffinePair, SelectedAffinePairLaunch, SelectedAffinePairSpec};
pub use reduce::{SelectedAffineReduce, SelectedAffineReduceLaunch, SelectedAffineReduceSpec};