fix: max_tokens 추가 — Gemma 16000, EXAONE 4096

응답이 중간에 끊기는 문제 해결. ModelAdapter에 max_tokens
파라미터 추가, stream/complete 양쪽 payload에 반영.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

nanoclaude/services/backend_registry.py

@@ -58,6 +58,7 @@ class BackendRegistry:
             system_prompt=REASONER_PROMPT,
             temperature=settings.reasoning_temperature,
             timeout=settings.reasoning_timeout,
+            max_tokens=16000,
         )
 
     def start_health_loop(self, interval: float = 30.0) -> None: