Bug concerning broadcast between device array and unified array

[mashu]

Describe the bug

This is isolated CUDA.jl broadcast example that fails when two arrays are of different type.

To reproduce

The Minimal Working Example (MWE) for this bug:

using CUDA
CUDA.rand(Float32,30,30) .+ cu(rand(Float32,30,30,1), unified=true)

Also initially posted FluxML/NNlib.jl#568 as I got it wrong how the dispatch behaves for .+ when dimensions are the same

CUDA 5.2.0

Expected behavior

Broadcast should produce an array, which it does in case of CPU CPU arrays.

rand(Float32,30,30) .+ rand(Float32,30,30,1)

Version info

Details on Julia:

Julia Version 1.10.2
Commit bd47eca2c8a (2024-03-01 10:14 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 32 × 13th Gen Intel(R) Core(TM) i9-13980HX
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, goldmont)
Threads: 1 default, 0 interactive, 1 GC (on 32 virtual cores)

Details on CUDA:

CUDA runtime 12.3, artifact installation
CUDA driver 12.0
NVIDIA driver 525.147.5

CUDA libraries: 
- CUBLAS: 12.3.4
- CURAND: 10.3.4
- CUFFT: 11.0.12
- CUSOLVER: 11.5.4
- CUSPARSE: 12.2.0
- CUPTI: 21.0.0
- NVML: 12.0.0+525.147.5

Julia packages: 
- CUDA: 5.2.0
- CUDA_Driver_jll: 0.7.0+1
- CUDA_Runtime_jll: 0.11.1+0

Toolchain:
- Julia: 1.10.2
- LLVM: 15.0.7

1 device:
  0: NVIDIA GeForce RTX 4090 Laptop GPU (sm_89, 14.758 GiB / 15.992 GiB available)