feat(translate): universal reasoning suppression for DeepSeek/Qwen/MiniMax/Kimi

Added disable_reasoning toggle that works via:
1. API parameter: reasoning_effort: 'none' (DeepSeek, Qwen)
2. System prompt instruction: 'Respond directly with ONLY the JSON.
   Do NOT include any reasoning, thinking, or chain-of-thought'
   (works for ALL models universally)

Backend: model column, schema, preview+executor payload
Frontend: checkbox in LLM settings, i18n en+ru

backend/src/models/translate.py

@@ -52,6 +52,7 @@ class TranslationJob(Base):
     target_languages = Column(JSON, nullable=True, comment="List of BCP-47 target language codes (multi-language support)")
     provider_id = Column(String, nullable=True, comment="LLM provider ID")
     batch_size = Column(Integer, nullable=False, default=50, comment="Records per batch")
+    disable_reasoning = Column(Boolean, default=False, comment="If true, pass reasoning_effort:none to LLM to save output tokens")
     upsert_strategy = Column(String, nullable=False, default="MERGE", comment="MERGE, INSERT, UPDATE")
 
     # Environment association