[GSK-2824] Difficulty level tests for rags

[pierlj]

• Add the possibility to store metadata for each question in the QATestset.
• Move difficulty as metadata.
• Add possibility to split a test suite into several test based on a metadata : testset.to_test_suite(slicing_metadata=["difficulty"]) will create a suite with as many tests as difficulty levels in the testset. Each test contains a filtered dataset with only one difficulty level.