wasi-nn: handle any kind of tensor layout

In #3956 we identified that OpenVINO's latest version has a new limitation on when the user must specify the layout of tensors. These layouts are somewhat image specific (e.g., `NHWC` = Number of images in a batch, Height in pixels, Width in pixels, number of Color channels) so their inclusion in the wasi-nn specification is dubious, since it should handle any kind of inference. OpenVINO expects some early configuration of this layout, though; currently the `NHWC` layout is forced as a sane default, but [other layouts are possible](https://docs.rs/openvino/latest/openvino/enum.Layout.html). To handle any kind of input layout, the OpenVINO backend could: 1) find another way of passing in the layout (i.e., upstream spec changes) or 2) avoid OpenVINO pre-processing that requires the layout (e.g., by using `ANY`).