Update all weights to the `.weights.h5` format

[mattdangerw]

We have a new "best practice" format for weight saving. We should re-upload all our existing checkpoints in this format.

Steps:

• Load a backbone with from_preset.
• Save in the new format weights by calling backbone.save_weights("model.weights.h5").
• On our gcp bucket. Make a new v2 directory keras_nlp/models/preset/xx_id/v2.
• Copy the new weights over along with all other preprocessing assets.
• Update our xx_preset.py file to point to the new assets.

Make sure not to delete of overwrite anything currently in the bucket! This should be non-destructive.

This will need to be handled by someone on the Keras team directly, as it requires write access to GCP storage.

[mattdangerw]

Done!