Remove manual fetching of NN model.json and weights.bin files and use tf.io.http instead

[lindapaiste]

Ref #113

• Figure out how to pass fetched data to TFJS using the IOHandler interface without converting to a File object.

Ref #84 (comment)

TBH this entire section of the code feels a bit odd and unnecessary. We are fetching the files, creating File objects, and then passing them to the TF model with tf.io.browserFiles. But TF is definitely capable of fetching files on its own so it seems like we shouldn't have to do that on our end. It seems like we should just be able to pass the URLs. That's a discussion for another time though, because maybe there's a reason that this is necessary and I'm just not seeing it.

Changes:

• Remove manual fetching of files via axios.get and conversion to File objects.
• Change tf.io.browserFiles to tf.io.http.
• Set the weightUrlConverter option to use the path from the user's object instead of the path that TFJS finds in weightsManifest section of the model.json file.

Note: weightUrlConverter is specified as an async function but I wrote it as a normal function. We can throw in an async keyword if there is concern.

I did try it out and it worked.

I even modified the path in the model.json file (to "paths":["./not-the-right-path/model.weights.bin"]) because in this example the override URL is the same as what's in the weightsManifest. We do this in order to support the p5.js Web Editor, where the actual file path is a remote Amazon S3 URL.

[shiffman]

Thank you @lindapaiste this seems like a much more concise way to load files, I tested all the examples and am going ahead to merge.