Deal with configurations without pre-binning

[marcverhagen]

Because

Models without any pre-binning were not dealt with before in the classifier/stitcher code, so it is not surprising that the stitcher breaks on this because it uses the pre-bin labels.

Done when

No response

Additional context

No response

[marcverhagen]

The classifier also breaks on this since it relies on there being pre-bins:

app-swt-detection/modeling/classify.py

Lines 52 to 56 in 973294b

	self.classifier = train.get_net(
	in_dim=self.featurizer.feature_vector_dim(),
	n_labels=len(self.model_config['bins']['pre'].keys()) + 1 if 'pre' in self.model_config['bins'] else 23,
	num_layers=self.model_config["num_layers"],
	dropout=self.model_config["dropouts"])