Add runtime config for model-specific jobs

[Tooyosi]

Make BaJoR accept model-coupled batch runtime config from job requests instead of relying only on deployment env vars.

This allows Kade to pass context runtime settings such as script paths, container image, promote script, and pretrained checkpoint reference while keeping BaJoR responsible for Azure Batch scheduling.

What changed

• Extended BaJoR job options to accept request-driven runtime fields:
  • container_image_name
  • training_script_path
  • prediction_script_path
  • promote_script_path
  • pretrained_checkpoint_url
• Added a runtime config resolver to centralize request-first, env-fallback behavior
• Updated finetuning scheduling to use request-driven:
  • training script path
  • pretrained checkpoint location
  • promote script path
  • container image
• Updated prediction scheduling to use request-driven:
  • prediction script path
  • checkpoint target
  • container image