Persist eval outputs for Dataface analysis and boards

Problem

CANCELLED — merged into "Set up eval leaderboard dft project and dashboards" task.

This task was originally framed as building an eval persistence layer. During planning, it became clear that the real deliverable is dashboards over JSONL, not persistence infrastructure. The leaderboard task covers the same scope more concretely. All context from this task has been incorporated there.

Possible Solutions

Plan

Implementation Progress

QA Exploration

• [x] QA exploration completed (or N/A for non-UI tasks)

N/A for browser QA initially. Once dashboards exist, verify they render correctly with sample eval output via dft serve.

Review Feedback

• [ ] Review cleared