evals: switch the default LLM judge to claude-haiku-4-5

Replaces gpt-5-mini-2025-08-07 as the default judge everywhere: the 147
eval configs plus the hardcoded defaults in the evals CLI, recipe composer,
comparison config generator, taubench user-simulator, trial runner and
OptimizeConfig.

Haiku is roughly 3x faster on the judging path and we are out of OpenAI
quota, so gpt-5-mini was failing closed on long eval runs.

tests/learning/test_optimize_cmd.py

@@ -113,7 +113,7 @@ class TestOptimizeConfig:
         assert cfg.max_trials == 20
         assert cfg.early_stop_patience == 5
         assert cfg.optimizer_model == "claude-sonnet-4-6"
-        assert cfg.judge_model == "gpt-5-mini-2025-08-07"
+        assert cfg.judge_model == "claude-haiku-4-5-20251001"
 
 
 class TestNewEventTypes: