Fix dropping of uninitialised memory in tensor0

[Ollie-Pearce]

tensor0 builds an uninitialised buffer and then writes to that value with an assignment operator. Assignment drops the uninitialised tensor before overwriting it with x.

let mut tensor = Tensor::uninitialized::<A>(&[]).unwrap();
tensor.as_slice_mut_unchecked::<A>()[0] = x;

Reproduction:

Test case:

#[test]
fn drop_of_uninit() {
    let _t: Tensor = tensor0(1.0f32);
}

Verifying with Miri:

cargo +nightly-2025-08-20 miri test -p tract-data --test repro

Miri output:

error: Undefined Behavior: constructing invalid value at .pointer: encountered 0, but expected something greater or equal to 1
    |
512 |         self.ptr.cast().as_non_null_ptr()
    |         ^^^^^^^^ Undefined Behavior occurred here
    |

Fix:

Replace the assignment operator with the std::ptr::write function which directly writes to x, fixing the drop of uninitialised memory.

[kali]

Hey, nice catch. My imaginary friend mention another possible occurences in data/src/tensor.rs:1440 and core/src/ops/array/range.rs:51 . Want to have a look at these too ?

[Ollie-Pearce]

Thanks, I'll take a look sometime over the next couple of days :)