feat: wire OTel telemetry spans into evaluation runner #90

Summary

Integrated the OTel telemetry module (telemetry.py) into services/runner.py to emit structured spans for every evaluation run.

What was done

• Added init_tracing() call at the start of run_evaluation()
• Wrapped the entire evaluation body in eval_run_span() context manager
• Added per-item eval_item_span() for each dataset row
• Added record_evaluator_span() after threshold evaluation for each evaluator
• Added set_eval_run_result() with pass_rate and item counts on the root span
• Added shutdown_tracing() at the end of the run
• Extracts model and agent_id from run_config.target.endpoint for span attributes

Span hierarchy

eval_run (root)
  eval_item[0]
    evaluator: SimilarityEvaluator
    evaluator: CoherenceEvaluator
  eval_item[1]
    ...

Implemented in

PR #86 (feature/telemetry-doc -> develop)