feat: Extend TransformersQueryClassifier: clean version

[anakin87]

Related Issue(s): fixes #2587

Proposed changes:

• clean version of Extend TransformersQueryClassifier #2903
• now TransformersQueryClassifiersupports text-classification and zero-shot-classification
• new tests

Pre-flight checklist

• I have read the contributors guidelines
• I have enabled actions on my fork
• If this is a code change, I added tests or updated existing ones
• If this is a code change, I updated the docstrings

Two main doubts:

• Does Fix validation for dynamic outgoing edges #2850 impact this PR? How?
• Should we provide an example of zero-shot-classification in Tutorial 14? (impacts Tutorial 14 is quite complicated to follow haystack-tutorials#34)

[ZanSara]

Does #2850 impact this PR? How?

It standardizes the way nodes can declare to have a variable number of outputs. Check out this method:

haystack/haystack/nodes/base.py

Line 131 in 91f954b

def _calculate_outgoing_edges(cls, component_params: Dict[str, Any]) -> int:

Right now you just declare 10 outgoing edges, as other nodes were doing, because there was no good way to tell the pipeline that your node could have a variable number of outputs. Now there is, so I recommend implementing that part in the new way.

You can have a look at FileTypeClassifier https://github.com/deepset-ai/haystack/blob/master/haystack/nodes/file_classifier/file_type.py or RouteDocuments https://github.com/deepset-ai/haystack/blob/master/haystack/nodes/other/route_documents.py to see how is it done. Documentation will follow 🙈

[ZanSara]

Should we provide an example of zero-shot-classification in Tutorial 14? (impacts deepset-ai/haystack-tutorials#34)

That would be fantastic! But maybe worth a separate PR. @TuanaCelik @mkkuemmel what do you think?

[mkkuemmel]

Should we provide an example of zero-shot-classification in Tutorial 14? (impacts deepset-ai/haystack-tutorials#34)

That would be fantastic! But maybe worth a separate PR. @TuanaCelik @mkkuemmel what do you think?

Yes, I like the idea! A small section at the bottom of Tutorial 14. We just need to make sure it remains understandable (but good markdown formatting would go a long way already). I do agree that a separate PR would be good for this.

[anakin87]

@ZanSara I've tried to implement the lightweight approach and move labels validation in _calculate_outgoing_edges.
WDYT?

[ZanSara]

Looks good! Thank you once again 😊

[ZanSara]

There has been another PR targeting this node with the goal of adding progress bars to the run method. I'm afraid I can't help too much with the merge, but it doesn't seem too messy either. Feel free to tag @vblagoje (PR author) if something is unclear or you need help

[anakin87]

I hope to have correctly fixed the conflict now.

@vblagoje 👋!
Related to #2864: in the run_batch method, I had to replace all_predictions.extend(predictions) with all_predictions.extend([predictions]) to avoid very unexpected effects (appending the keys instead of the whole prediction dictionary), which made my new tests fail; probably, the CI didn't show errors because the Query Classifier was largely uncovered by tests...