refactor: Change output of _find_ml_task from "unsupported" to "unknown"

[augustebaum]

As discussed with @glemaitre, "unsupported" makes it seem like skore is not usable in certain cases (e.g. using an EstimatorReport with a Keras model), whereas it just makes it more difficult for us to add extra information (e.g. in the EstimatorReport, to suggest specific metrics to compute).