Replace model specific score functions with generic ones?

[RemyLau]

Currently, each model comes with a .score function for evaluating the model performance. Most of them do the same thing, e.g., compute model accuracy for cell-type annotation tasks. Why not just have one generic function that calculates the accuracy (and similarly for other tasks)?

[RemyLau]

• Follow scklearn convention: score takes feature and label as input and spits out score, with an option to select the type of score.
• Refactor models to make generic score function

[RemyLau]

Closing as it is being tracked in #143