Support loading models from ModelArtifacts and ModelJSON in loadGraphModelSync

[mattsoulanille]

Let users pass inputs of type ModelArtifacts or [ModelJSON, ArrayBuffer] to loadGraphModelSync instead of requiring them to construct a synchronous IOHandler. In the case of ModelArtifacts, the model weights are contained in the object. In the case of [ModelJSON, ArrayBuffer], the ArrayBuffer is a list of concatenated weights for the model.

This change turns this code:

const modelJson = JSON.parse(modelString);
const weights = ...
modelJson.weightData = weights;
modelJson.weightSpecs = modelJson.weightsManifest[0].weights;
const ioHandler = tf.io.fromMemorySync(modelJson);
const model = tf.loadGraphModelSync(ioHandler);

... into this simpler code

const modelJson = JSON.parse(modelString);
const weights = ...
const model = tf.loadGraphModelSync([modelJson, weights]);

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

---

This change is 

[ahmedsabie]

Reviewed 2 of 2 files at r1.
Reviewable status: complete! 2 of 1 approvals obtained