burn-cubecl-fusion: missing FloatKind::F64 arm causes panic with f64 precision on CPU backend

**Describe the bug** Using burn-cpu (CubeCL CPU backend) with Cpu<f64, i32> and fusion enabled (default features) panics at runtime with "Unsupported precision for fusion: f64". The panic occurs in burn-cubecl-fusion-0.21.0/src/engine/codegen/ir.rs line 787, in the From<ElemType> for FuseType implementation. The FloatKind::F64 arm is missing from the match, even though FuseType::F64 is already defined (line 579) and the reverse conversion FuseType::F64 → ElemType::Float(FloatKind::F64) exists (line 828). Version: burn 0.21.0 **To reproduce** ```rust // Cargo.toml // burn = { version = "0.21.0", features = ["std", "cpu", "autodiff"] } use burn::backend::Autodiff; use burn_cpu::{Cpu, CpuDevice}; use burn::tensor::{Tensor, TensorData, DType}; type MyBackend = Autodiff<Cpu<f64, i32>>; fn main() { let device = CpuDevice::default(); let t = Tensor::<MyBackend, 1>::from_data( TensorData::from(&[1.0_f64, 2.0, 3.0][..]), (&device, DType::F64), ) .require_grad(); let out = t.clone().powf_scalar(2.0).sum(); let grads = out.backward(); let grad = t.grad(&grads).unwrap(); println!("{:?}", grad.to_data()); } ``` Panics on the backward pass with: thread 'DSU-0-0' panicked at burn-cubecl-fusion-0.21.0/src/engine/codegen/ir.rs:787:22: Unsupported precision for fusion: f64 **Suggested Fix** Add the missing FloatKind::F64 => Self::F64 arm in burn-cubecl-fusion/src/engine/codegen/ir.rs, line 786: ```rust ElemType::Float(kind) => match kind { FloatKind::F16 => Self::F16, FloatKind::BF16 => Self::BF16, FloatKind::F32 => Self::F32, FloatKind::F64 => Self::F64, // add this FloatKind::Flex32 => Self::Flex32, _ => panic!("Unsupported precision for fusion: {value}"), }, ```