CuSparseMatrxCSC{<:Complex}*CuMatrix{<:Real} triggers allowscalar error

**Describe the bug** Multiplication of complex CuSparseMatrixCSC and real dense CuMatrix (and vice-versa) fail with `ERROR: LoadError: Scalar indexing is disallowed.`. One would expect the multiplication to proceed, and a resulting complex dense CuMatrix to come out. **To reproduce** The Minimal Working Example (MWE) for this bug: ```julia using CUDA using CUDA.cuSPARSE using LinearAlgebra using SparseArrays A = CuSparseMatrixCSC(sprand(Float64, 100, 100, 0.01)) B = CUDA.randn(Float64, 100, 100) C = A * B # works A = CuSparseMatrixCSC(sprand(ComplexF64, 100, 100, 0.01)) B = CUDA.randn(Float64, 100, 100) C = A * B # does not work ``` Note: the issue occurs with AMDGPU too. I will link the relevant issue once open. Edit: https://github.com/JuliaGPU/AMDGPU.jl/issues/904 **CUDA.versioninfo():** ``` CUDA toolchain: - runtime 13.1, artifact installation - driver 590.48.1 for 13.2 - compiler 13.2 CUDA libraries: - cuBLAS: 13.2.1 - cuSPARSE: 12.7.3 - cuSOLVER: 12.0.9 - cuFFT: 12.1.0 - cuRAND: 10.4.1 - CUPTI: 2025.4.1 (API 13.1.1) - NVML: 13.0.0+590.48.1 Julia packages: - CUDACore: 6.0.0 - GPUArrays: 11.5.3 - GPUCompiler: 1.9.1 - KernelAbstractions: 0.9.41 - CUDA_Driver_jll: 13.2.1+0 - CUDA_Compiler_jll: 0.4.3+0 - CUDA_Runtime_jll: 0.21.0+1 Toolchain: - Julia: 1.12.6 - LLVM: 18.1.7 Preferences: - CUDA_Runtime_jll.version: 13.1 4 devices: 0: NVIDIA GH200 120GB (sm_90, 94.347 GiB / 95.577 GiB available) 1: NVIDIA GH200 120GB (sm_90, 94.995 GiB / 95.577 GiB available) 2: NVIDIA GH200 120GB (sm_90, 94.995 GiB / 95.577 GiB available) 3: NVIDIA GH200 120GB (sm_90, 94.993 GiB / 95.577 GiB available) ```