Testset generator not preserving persona and scenario metadata in generated samples

[anistark]

When using TestsetGenerator with multiple personas, there's no reliable way to track which persona was used to generate each sample in the resulting testset. The persona information is lost after generation, making it impossible to:

• Correlate generated questions with their source personas
• Verify persona distribution in the generated dataset
• Validate that questions are relevant to their intended personas

Users currently resort to workarounds like string prefixing or processing one persona at a time, which is inefficient and degrades quality. A recommended approach to preserve and access persona metadata in generated samples is needed.

[dev-jonathan]

I've created PR #2389 that adds metadata preservation for SingleHop query generation. The PR adds:

• persona_name
• query_style
• query_length

These fields are automatically populated during sample generation in SingleHopQuerySynthesizer, solving the metadata tracking issue for single-hop scenarios. Once merged, it will allow proper tracking and analysis of persona distribution in generated samples.

I hope it is useful!