MLTensor explainer: replace "import buffer" with "export tensor"

[bbernhar]

Proposes an IDL change to reflect our current implementation for WebGPU interop. Specifically, it moves the responsibility of producing a interop GPUBuffer into WebNN.

Before

dictionary GPUImportedTensorDescriptor : GPUObjectDescriptorBase {
  required MLTensor source;
};

partial interface GPUDevice {
  Promise<GPUBuffer> importExternalBuffer(GPUImportedTensorDescriptor descriptor);
}

After

partial interface MLContext {
  Promise<GPUBuffer> exportToWebGPU(MLTensor source);
}

This change achieves the following:

• Eliminates the need to modify (or extend) the WebGPU spec for an experimental WebNN feature.
• Gives WebNN WG flexibility to iterate/extend the interop API along with the WebNN IDL (ex. new sources).

[bbernhar]

@reillyeon @huningxin @RafaelCintron will follow up with the IDL changes but I figured we start with updating the explainer

If no blocking concerns ==👍

[bbernhar]

@huningxin Feel free to merge - will follow-up with IDL changes.

[fdwr]

👍