Re-enable reasoning for accurate verdicts (no-think rubber-stamped 'match')

Disabling thinking made reasoning models mark everything 'match' even when ref/gen
clearly differ. Added an enable_thinking toggle (default ON) threaded through the
generation path; the prompt now allows reasoning then asks for the result, and
verdict_rule explicitly warns against lazy 'match'. _parse_json now scans for the
JSON object AFTER the reasoning prose (last balanced object with 'axes'), and the
markdown fallback already reads reasoned per-axis output. Default max_new_tokens
2048->3072 so verdicts don't get cut off.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

workflow/workflow_api.json

@@ -68,7 +68,7 @@
       "model_path": "/media/p5/qwen3vl_4b_abliterated_comfy_convert/hf_bf16",
       "precision": "bf16",
       "profile": "general",
-      "max_new_tokens": 2048,
+      "max_new_tokens": 3072,
       "temperature": 0.0,
       "swap_eval": true,
       "keep_loaded": true,

workflow/workflow_describe_api.json

@@ -12,7 +12,7 @@
       "profile": "general",
       "model_path": "/media/p5/qwen3vl_4b_abliterated_comfy_convert/hf_bf16",
       "precision": "bf16",
-      "max_new_tokens": 2048,
+      "max_new_tokens": 3072,
       "temperature": 0.0,
       "swap_eval": false,
       "keep_loaded": true,