fix(backend): restore the large-body limit on the voice route

POST /voice/extract returned 500 for any real recording. The threshold was
exactly 100 kb — Express's body-parser default — which is about 20 seconds of
audio, so the endpoint was unusable at its own 2-minute cap.

The scoped parser was registered as a path-mounted json() stacked in front of a
default one, which relied on two implicit behaviours: Express stripping the
mount path, and body-parser skipping a request another parser had already
handled. That coupling broke when the surrounding middleware order shifted, and
it broke silently — the parser was still registered, just no longer the one that
ran. Bisected by dumping the Express layer stack and confirming the raw error was
`entity.too.large` with `limit: 102400`.

Replaced with a single middleware that picks a parser by path. No mount-path
stripping, no dependence on parser ordering. Extracted to common/body-parsers.ts
so it is covered by a unit test rather than only reachable through main.ts, which
createTestingModule never executes.

The test is mutation-checked: forcing the default parser fails 2 of its 5 cases.
It also pins that the larger limit does not leak app-wide, and that a merely
similar path (/api/voice/extract/extra) does not get it.

Verified against the compiled server: 300 kb now reaches /api/voice/extract,
/api/auth/login still rejects it, and ordinary requests are unaffected.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

backend/src/main.ts

@@ -1,7 +1,8 @@
 // backend/src/main.ts
 import { NestFactory } from '@nestjs/core';
-import { json, urlencoded } from 'express';
+import { urlencoded } from 'express';
 import { AppModule } from './app.module';
+import { createJsonBodyParser } from './common/body-parsers';
 import { ValidationPipe } from '@nestjs/common';
 import cookieParser from 'cookie-parser'; // 👈 Change this line!
 import { DocumentBuilder, SwaggerModule } from '@nestjs/swagger';
@@ -29,12 +30,8 @@ async function bootstrap() {
   // reject a voice recording at its 100 kb default before any later middleware ran.
   const app = await NestFactory.create(AppModule, { bodyParser: false });
 
-  // Voice recordings are base64 JSON and pass 100 kb at roughly 20 seconds of audio.
-  // Registered first and scoped to the one route: body-parser marks the request handled,
-  // so the default-limit parser below skips it and every other endpoint keeps the
-  // standard limit.
-  app.use('/api/voice/extract', json({ limit: '10mb' }));
-  app.use(json());
+  // Voice needs a larger JSON limit than everything else; see body-parsers.ts.
+  app.use(createJsonBodyParser());
   app.use(urlencoded({ extended: true }));
   
   app.useGlobalFilters(new HttpExceptionFilter());