Fix for weighted BCE loss

[aditya0by0]

Fixes #131

Changes

• Argument Linking for Weighted BCE Loss,
• Raise error if no weights given for Weighted BCE loss
• Another Class for Unweighted BCE Loss
• Update MLP test for Unweighted BCE Loss

[aditya0by0]

The below were changes deleted for weighted bce link. But I am just reverting only the first argument linking as the second one serves no purpose.

        parser.link_arguments(
            "data", "model.init_args.criterion.init_args.data_extractor"
        )
        parser.link_arguments(
            "data.init_args.chebi_version",
            "model.init_args.criterion.init_args.data_extractor.init_args.chebi_version",
        )

https://github.com/ChEB-AI/python-chebai/pull/71/files

[sfluegel05]

I like the separation of weighted / unweighted into separate classes. If we force the user to use the beta parameter, why not do that in the signature? So instead of having beta: Optional[float] = None, we could have beta: float = 0.99. In my experience, 0.99 is the only value I have ever used there so it would be a sensible default.

[sfluegel05]

Thanks for implementing this.