select model for inserting slices according the model id

[lpi-tn]

This pull request introduces several changes to improve the handling of embedding models and their corresponding Qdrant collections in the qdrant_handler module and associated test files. The most significant updates include transitioning from language-specific collection names to embedding model-specific naming, refactoring the classify_documents_per_collection function for clarity and efficiency, and updating the test cases to align with these changes.

Updates to embedding model handling:

• Modified the classify_documents_per_collection function in welearn_datastack/modules/qdrant_handler.py to use a mapping (model_name_collection_name) for resolving collections based on embedding model names, replacing the previous language-based logic. This simplifies the logic and reduces redundancy.

Test case updates for embedding model-specific collections:

• Updated collection names in tests/qdrant_syncronizer/test_qdrant_handler.py and tests/qdrant_syncronizer/test_qdrant_syncronizer.py to follow the new embedding model-specific naming convention (e.g., collection_welearn_en_embmodel → collection_welearn_en_english-embmodel). [1] [2]
• Refactored test cases in test_qdrant_handler.py to include the embedding model name as a parameter for FakeSlice instances, ensuring proper alignment with the new naming logic. [1] [2] [3]
• Introduced an EmbeddingModel instance in test_qdrant_syncronizer.py to simulate embedding model metadata and updated DocumentSlice instances to reference this model. [1] [2]

Additional improvements:

• Added defaultdict import in welearn_datastack/modules/qdrant_handler.py to simplify the initialization of the ret dictionary in classify_documents_per_collection.
• Incorporated Mock and EmbeddingModel imports in test files to support the new embedding model logic. [1] [2]

[Copilot]

Pull Request Overview

This PR refactors the collection mapping logic in the qdrant_handler module and updates the related test cases to use embedding model-specific naming instead of language‐based collection names. Key changes include:

• Replacing manual dictionary initialization with a defaultdict(set) in classify_documents_per_collection.
• Creating a mapping from embedding model title to collection name based on the fourth component of the collection name.
• Updating test cases in qdrant_syncronizer and qdrant_handler to use the new naming convention.

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
welearn_datastack/modules/qdrant_handler.py	Refactored classification logic with new mapping.
tests/qdrant_syncronizer/test_qdrant_syncronizer.py	Updated collection names and test setup.
tests/qdrant_syncronizer/test_qdrant_handler.py	Updated FakeSlice usage and expected results.

Comments suppressed due to low confidence (2)

tests/qdrant_syncronizer/test_qdrant_handler.py:134

• The expected collection name for the multilingual case ('collection_welearn_mul_mulembmodel') does not align with the new mapping using the embedding model title; either update the test expectation or adjust the classification function to implement the intended fallback behavior.

            "collection_welearn_mul_mulembmodel": {doc_id1},

welearn_datastack/modules/qdrant_handler.py:48

• With the removal of the fallback logic, slices that do not match any key in the mapping are simply skipped; if multilingual fallback is intended (as suggested by test expectations), consider reintroducing logic to derive the collection name from the document's language.

        except KeyError: