Find better strategies to generate values for `weights` and `bias` tensors

Bias and Weights tensors are currently more or less ignored by the strategy, using respectively the very scientific values of 0.0 and 1.0 for all elements. Unfortunately they kind of stand out in the AIDGE world of tensors (they are not explicit inputs, are set and dimensioned upon creation of an operator, and exist directly as tensors not np.array).