fix: wire crewai worker to top-level models

The worker still depended on the removed chat/embedding provider
services, and no service was attached to the new top-level models, so
the CHAT_URL/CHAT_MODEL/EMBEDDING_URL/EMBEDDING_MODEL env vars read by
app/core/ai_clients.py were never injected. Attach both models to the
worker via endpoint_var/model_var (compose.local.yaml inherits this
through extends), rename the chat model back to match the local file,
and drop x-defang-llm to match the other samples on the new syntax.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sw7j9FYnbChZJbYArxbhuo

Author: defangdevs

samples/crewai/compose.yaml

@@ -65,24 +65,27 @@ services:
     depends_on:
       - postgres
       - redis
-      - chat
-      - embedding
     environment:
       DJANGO_SETTINGS_MODULE: config.settings
       DATABASE_URL: postgres://postgres:${POSTGRES_PASSWORD}@postgres:5432/postgres?sslmode=require # remove sslmode for Playground
       REDIS_URL: redis://redis:6379/0
       OPENAI_API_KEY: defang
       DJANGO_SECRET_KEY: null
+    models:
+      chat:
+        endpoint_var: CHAT_URL
+        model_var: CHAT_MODEL
+      embedding:
+        endpoint_var: EMBEDDING_URL
+        model_var: EMBEDDING_MODEL
     deploy:
       resources:
         reservations:
           cpus: "0.5"
           memory: 512M
 
 models:
-  llm:
+  chat:
     model: chat-default
-    x-defang-llm: true
   embedding:
     model: embedding-default
-    x-defang-llm: true