Compare pose-detection prediction data with another GPU backend

[gyagp]

We couldn't get CPU prediction data for pose-detection. So to compare the result, we use another GPU backend for reference.

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

---

This change is 

[gyagp]

@haoyunfeix PTAL

[gyagp]

@Linchenn If pose-detection will be removed very soon, we don't need this patch. Otherwise, we hope to bring this in temporarily so that our daily test can pass.

[Linchenn]

Thank you Yang! I have reached a consensus over the new model list with Na, Ping and Jiajia (I have pinged you on the spreadsheet), so I will update the model.json soon.

Even though the failure will disappear when the model is removed, I am curious about if WebGL/WebGPU's results are expected or CPU/WASM's results and hope we could fix the issue.

I took a look over the correctness check for pose-detection and the following codes returns an empty array:

<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs"></script>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow-models/toxicity"></script>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow-models/pose-detection"></script>
<script>

async function loadImage(imagePath) {
    const image = new Image();
    const promise = new Promise((resolve, reject) => {
      image.crossOrigin = '';
      image.onload = () => {
        resolve(image);
      };
    });
  
    image.src = imagePath;
    return promise;
}

async function run() {
    const image = await loadImage('https://storage.googleapis.com/tfjs-models/assets/posenet/tennis_standing.jpg');
    const runtime = 'tfjs';
    const enableSmoothing = false;
    const modelType = 'lite';
    config = {runtime, enableSmoothing, modelType};
    model = await poseDetection.createDetector('BlazePose', config);
    tf.setBackend('webgl');
    let res = await model.detectPose(image);
    console.log(res);
}

run();
</script>

@ahmedsabie Could you help take a look at the above codes? Is empty array expected? Or did I miss something? (@lina128 is OOO this week)

By the way, I found the issue exists when adding pose-detection, https://github.com/tensorflow/tfjs/pull/5147

[gyagp]

Talked with Jiajia about new benchmarks, and she will leave some more comments in the spreadsheet. I will close this PR once the new benchmarks are merged.