Improve observability: add structured logging to tournament, analyze, and verify_parity

[lukeinglis]

Closes #157

Changes

• tournament.py — Added structlog logger and logger.info() calls to all public functions: load_tournament_results, model_source_matrix, paired_bootstrap_test, pairwise_comparison_table, cost_accuracy_summary, tournament_report
• analyze.py — Added logger.info() calls to analysis functions missing them: analyze_by_source, analyze_calibration, analyze_biases, analyze_decomposition, analyze_worst_questions, analyze_by_horizon, compare_paired, save_analysis
• verify_parity.py — Added structlog logger and logger.info() calls to all check functions: check_prompt_templates, check_resolution_matching, check_scoring_formula, check_missing_forecast_default, check_multi_horizon_batching, check_question_count, check_dummy_score, check_resolution_outcome_diversity, check_resolution_entry_preservation, check_cross_round_filtering, check_score_comparison, check_per_source_breakdown
• eval.py — Added logger.info() calls to save_result, load_previous_results, split_held_out, list_rounds, print_leaderboard_comparison
• lab_forecaster.py — Added logger.warning() to _parse_probability failure path and logger.info()/logger.warning() to _extract_probabilities
• tests/test_logging_config.py — Extended smoke test to cover newly instrumented modules (tournament, analyze, verify_parity)

[lukeinglis]

Duplicate of #158 — both candidates independently picked the same logging backlog item during outer loop generation 0. Closing in favor of #158.