perf(server): Switch spell/translate proxy to gpt-5-nano

Halves input cost ($0.10→$0.05/M) vs gpt-4o-mini and improves
quality. Drop temperature=0.2 (gpt-5 only accepts default).

Author: gido

server/.env.example

@@ -1,4 +1,5 @@
 PORT=3000
 OPENAI_API_KEY=sk-...
+OPENAI_MODEL=gpt-5-nano
 ALLOWED_ORIGINS=http://localhost:3000,https://antistatique.harvestapp.com,https://*.harvestapp.com,moz-extension://*,chrome-extension://*
 ESTI_SECRET=pat-... # Re-use your Airtable PAT here (or leave empty in dev)

server/routes/spelling.js

@@ -6,7 +6,7 @@ const router = Router();
 
 const openaiKey = process.env.OPENAI_API_KEY;
 const openai = openaiKey ? new OpenAI({ apiKey: openaiKey }) : null;
-const MODEL = process.env.OPENAI_MODEL || 'gpt-4o-mini';
+const MODEL = process.env.OPENAI_MODEL || 'gpt-5-nano';
 
 // Very naive local fallback to allow basic testing without OpenAI
 function localSpellcheck(text) {
@@ -92,7 +92,6 @@ router.post('/', async (req, res) => {
           { role: 'system', content: system },
           { role: 'user', content: user },
         ],
-        temperature: 0.2,
         response_format: { type: 'json_object' }
       });
     } catch (apiErr) {

server/routes/translate.js

@@ -6,7 +6,7 @@ const router = Router();
 
 const openaiKey = process.env.OPENAI_API_KEY;
 const openai = openaiKey ? new OpenAI({ apiKey: openaiKey }) : null;
-const MODEL = process.env.OPENAI_MODEL || 'gpt-4o-mini';
+const MODEL = process.env.OPENAI_MODEL || 'gpt-5-nano';
 
 router.post('/', async (req, res) => {
   try {
@@ -59,7 +59,6 @@ router.post('/', async (req, res) => {
           { role: 'system', content: system },
           { role: 'user', content: user },
         ],
-        temperature: 0.2,
         response_format: { type: 'json_object' }
       });
     } catch (apiErr) {