Memory leak when running GPU broadcast in a loop

I need to run GPU operations inside a loop, where the output of one iteration is used in the next one.

However, even very simple GPU broadcasts result in a memory leak and eventually I get OutOfMemoryError().

I am using oneAPI v.1.2.2 on WSL2 with Ubuntu on a Windows 10.

A very simple example that reproduces this is:

```
using oneAPI

gpu_array = oneAPI.zeros(Float32, 10_000_000)

for j in 1:5_000
gpu_array .+= 1
end
```

Is there something I am missing here?

I can see the GPU memory fill up in the Task Manager:

Start by running the provided Julia reproducer on oneAPI 1.2.2 and observing GPU memory during repeated `.+=` operations. Trace the GPU broadcast path and its allocation or lifetime behavior in oneAPI.jl. Done means the loop no longer continually increases GPU memory until OutOfMemoryError().

Written by the indexing model from the issue text.