Config: evalscope generation_config format + our extras (repeats, env) at bench level; parallel_tool_calls noted as unsupported in adapter

Co-Authored-By: Claude <noreply@anthropic.com>

evalharness/config/dp4-nothink.yaml

@@ -16,12 +16,14 @@ aime24:
     top_p: 1.0
     stream: true
     max_tokens: 32768
+  repeats: 12
 aime25:
   generation_config:
     temperature: 1.0
     top_p: 1.0
     stream: true
     max_tokens: 32768
+  repeats: 12
 mmlu_pro:
   generation_config:
     temperature: 0.0
@@ -61,12 +63,14 @@ aime26:
     top_p: 1.0
     stream: true
     max_tokens: 32768
+  repeats: 12
 hmmt26:
   generation_config:
     temperature: 1.0
     top_p: 1.0
     stream: true
     max_tokens: 32768
+  repeats: 12
 imo_answerbench:
   generation_config:
     temperature: 1.0
@@ -164,10 +168,8 @@ tau2_bench:
       generation_config:
         temperature: 0.0
         max_tokens: 4096
-  agent_config:
-    mode: native
-    strategy: react
-    max_steps: 50
+  env: tau2_official
+  max_turns: 50
 general_fc:
   generation_config:
     temperature: 0.0
@@ -185,6 +187,7 @@ bfcl_v3:
     extra_params:
       is_fc_model: true
       underscore_to_dot: true
+  env: bfcl_mock
 swe_bench_verified:
   generation_config:
     temperature: 0.0