Accessing a matrix via boolean mask doesn't work

MWE:
```julia
using OpenCL, pocl_jll
x12 = OpenCL.rand(10)
x13 = OpenCL.rand(Bool, 10)
x12[x13]
```

Fails with:
```
BoundsError: attempt to access 10-element OpenCL.CLArray{Float32, 1, OpenCL.cl.UnifiedDeviceMemory} at index [10-element OpenCL.CLArray{Bool, 1, OpenCL.cl.UnifiedDeviceMemory}]
```

Its CUDA alternative works as desired.