Run zero-shot baseline on selected models and question set

[eemilkos]

Use the implemented evaluation pipeline to run zero-shot prompts on the selected model set using the prepared question set.

This issue is about execution, not infrastructure design.

The run should specify:

• which models are included
• which question set version is used
• which prompt template is used
• where outputs are stored

The goal is to produce the first baseline dataset for later scoring and analysis.

Acceptance criteria

• The selected models for this run are listed.
• The question set version used for the run is identified.
• Zero-shot prompts are executed on all selected models or a clearly stated subset.
• Outputs are stored in the agreed structured format.
• Any failed requests are preserved in the logs rather than silently dropped.
• A short result summary is added to the issue or repository notes.