Evaluation: Optimize fetch results

[AkhileshNegi]

Is your feature request related to a problem?

When evaluation results are completed, we loop through each trace and store the results in the database. This is resource-intensive on the first fetch. After that, we read from the database to show data. Storing long results per trace bloats the database over time.

Describe the solution you'd like

Store evaluation results in S3 instead of the database:

• Save results to S3 when evaluation completes
• Read from S3 when displaying results
• Reduces database bloat from storing lengthy per-trace results

Additional consideration

Check if Langfuse SDK v3 provides an easier way to fetch evaluation results at the trace level - this may influence whether moving to S3 is still the right approach.