AGNO pipeline prompt refinement

[sshenzha]

Problem: Currently, we have tested the LLM-as-a-judge for SQL query equivalence only on 56 queries (one query derived from each template). This has given us an average of 57%. In some reasoning, the judge LLM assigned a lower score because the generated query contained words like 'LIMIT 1000', or for concept relationships, the generated query used JOINS instead of 'maps to'.

Solution: Modify the database agent to remove the LIMIT keyword and to use for concept_relationships the 'maps to' approach, which is the way that the ground truth was crafted