AveragedPerceptron factory method uses wrong default loss function

[KirillShlenskiy]

System Information:

• OS & Version: Windows 11
• ML.NET Version: 3.0.0-preview.23266.6
• .NET Version: .NET 6.0

Describe the bug StandardTrainersCatalog.AveragedPerceptron factory method uses LogLoss as its default loss function, which contradicts method documentation and AveragedPerceptronTrainer.Options defaults.

AveragedPerceptron method summary states:

<param name="lossFunction">The loss function minimized in the training process. If null, HingeLoss would be used and lead to a max-margin averaged perceptron trainer.</param>

This results in the following inconsistency in behaviour:

// Uses LogLoss:
MLContext.BinaryClassification.Trainers.AveragedPerceptron("Label");

// Uses HingeLoss:
MLContext.BinaryClassification.Trainers.AveragedPerceptron(new AveragedPerceptronTrainer.Options { LabelColumnName = "Label" });

Expected behavior 2 options:

• Make the non-options AveragedPerceptron factory method overload use HingeLoss as its default loss function (breaking change).
• Amend the method documentation.