config: remove hardcoded defaults, add nvidia_enable_thinking setting

All LLM parameters must now be explicitly set in .env — no silent
fallbacks in code. Added nvidia_enable_thinking bool and wired it into
the chat completion extra_body instead of the previous hardcoded True.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

server/app/main.py

@@ -68,7 +68,7 @@ async def generate_recipes(request: RecipeRequest):
                 top_p=settings.nvidia_top_p,
                 max_tokens=settings.nvidia_max_tokens,
                 extra_body={
-                    "chat_template_kwargs": {"enable_thinking": True},
+                    "chat_template_kwargs": {"enable_thinking": settings.nvidia_enable_thinking},
                     "reasoning_budget": settings.nvidia_reasoning_budget,
                 },
                 stream=True,