Evaluation: STT Config

[AkhileshNegi]

Is your feature request related to a problem?

In the first cut of STT evaluation, runs are created with hardcoded parameters (run_name, dataset_id, models). There is no way to pass a standardized config, making it difficult to manage and reproduce evaluation runs with different configurations (e.g. model settings, evaluation parameters). Text evaluation already solves this with config_id + config_version.

Describe the solution you'd like

Add config support to STT evaluation, following the text evaluation pattern:

• Accept config_id and config_version as input parameters in STTEvaluationRunCreate
• Resolve the config blob using ConfigVersionCrud before starting the evaluation batch
• Store config_id and config_version in the EvaluationRun record
• Extract STT-specific parameters (models, evaluation settings, etc.) from the resolved config