improve: use decorators rather than wrapper classes

[carynbear]

The issue with AlgorithmChecker and SystemChecker is that methods/attributes are not forwarded to the wrapped class which reduces the flexibility of DynaDojo.

For example, you cannot access an underlying Tensorflow or Pytorch neural net model without first calling ._alg which means you can't move the model to GPU natively.

I think we should migrate to using decorators.

Relevant stackoverflow post: https://stackoverflow.com/a/63644860