Warn about frozen weights use in `TensorflowModel` and `TorchModel`.

Related to issue #15 (closed)

This MR does not fix the aforementioned issue, that will require a minor extension to the current Model API, that will be accompanied by an extension of the current test scenarios.

Instead, it adds temporary warnings in TensorflowModel and TorchModel that are triggered when a wrapped model has frozen weights (i.e. parameters that do not require updating via gradient-descent). These warnings document the known limitations of the current support for such models (namely, that weight decay and Regularizer plug-ins cannot be used with them) and point towards the (future) version 2.1.0, that will fix those.