fix: Qwen3 defaults to enable_thinking=False — was returning empty answers

Root cause of 'arborist abstains on everything with qwen' (fox 2026-05-21):
Qwen3 thinking-on default burns the entire token budget on hidden <think>
reasoning over a 20K RAG context and returns EMPTY message.content
(measured: 768/768 completion tokens, content '') -> every arborist answer
UNGROUNDED. Bench harnesses passed enable_thinking=False via the MODELS
dict, but the CLI + control_ab did not, so the quality bench was measuring
a thinking-budget-exhaustion artifact, not abstention.

OpenAICompatibleClient now defaults Qwen3 to enable_thinking=False unless a
caller set it explicitly (reasoning-variant path passes True, preserved).
Verified: same France query goes empty/UNGROUNDED -> STRICT 'Nicolas
Sarkozy' with the flag. Fixes every caller (CLI, control_ab). 5 tests;
full suite 2547 passed. Today's qwen QUALITY numbers are void and need
re-running; energy numbers stand (real inference happened regardless).

tests/test_qwen_thinking_default.py

@@ -0,0 +1,34 @@
+"""Qwen3 enable_thinking=False default (2026-05-21).
+
+Qwen3 thinking-on default burns the token budget on hidden <think> over a
+large RAG context and returns empty content → every answer UNGROUNDED.
+The client forces enable_thinking=False for Qwen3 unless a caller set it.
+"""
+from arborist.qa.client import _qwen3_no_think_default as f
+
+
+def test_qwen3_gets_no_think_by_default():
+    eb = f("Qwen3.6-27B-UD-Q4_K_XL.gguf", None)
+    assert eb["chat_template_kwargs"]["enable_thinking"] is False
+
+
+def test_qwen3_merges_into_existing_extra_body():
+    eb = f("Qwen3.6-27B", {"guided_json": {"x": 1}})
+    assert eb["guided_json"] == {"x": 1}
+    assert eb["chat_template_kwargs"]["enable_thinking"] is False
+
+
+def test_explicit_thinking_true_is_preserved():
+    # the reasoning-variant path must keep thinking ON.
+    eb = f("Qwen3.6-27B", {"chat_template_kwargs": {"enable_thinking": True}})
+    assert eb["chat_template_kwargs"]["enable_thinking"] is True
+
+
+def test_non_qwen_untouched():
+    assert f("adamo1139/Hermes-3-Llama-3.1-8B-FP8-Dynamic", None) is None
+    eb = {"foo": 1}
+    assert f("Hermes-3", eb) is eb  # unchanged object
+
+
+def test_empty_model_untouched():
+    assert f("", None) is None