[Fix] TF Backend Regularization fix.

[agniv-the-marker]

TF Regularization

Check the docs and the argument for tf.keras.regualizers.l1/l2 is of the form l1/l2 instead of just l in the current version. Without it this breaks:

net = dde.maps.PFNN(
    network,
    "swish",
    "Glorot normal",
    regularization=["l2", 1e-8]
)

Specifically it gives you the error:

TypeError: L1.__init__() got an unexpected keyword argument 'l'

As of right now, the following works instead:

net = dde.maps.PFNN(
    network,
    "swish",
    "Glorot normal",
    regularization=["l1+l2", 1e-8, 1e-8]
)

But merging would allow for each one to be used on their own.

[lululxvi]

Split the regularization and model save into two PRs.