LLM-as-a-judge: use same model list as prompts

[mmabrouk]

Problem

The LLM-as-a-judge evaluator has a hardcoded list of models in its settings template (multiple_choice type with static options). Meanwhile, prompts in the playground use the dynamic model list from the model registry / vault.

Users expect to use the same models for evaluation that they use for their prompts.

Blocked By

This is blocked by the evaluator playground migration (AGE-3656). The fix requires rethinking how evaluator schemas work — specifically, the model field in LLM-as-a-judge should reference the same dynamic model list instead of being a hardcoded multiple_choice in the evaluator catalogue.

Notes from Sprint Planning (Mar 11)

• Related to the new schema design for evaluators
• Should be addressed as part of the evaluator playground migration work

[linear]

AGE-3707