Rethink how we do masking

[pstjohn]

It might be nice to not have to assume a mask value for the users input, we could perhaps add an additional field to policy_inputs 'mask': 1, that gets padded with zeros.

rlmolecule/rlmolecule/alphazero/tf_keras_policy.py

Line 98 in a2c6e68

batched_inputs_with_mask += [layers.Masking(mask_dict[input_name])(batched_input)]