Fase 0: estructura base de carpetas y docker-compose de training

Scaffolding inicial del repo: carpetas scripts/, data/schemas/,
data/raw/, out/, .gitignore para binarios/checkpoints, y el
docker-compose.yml del contenedor de training (imagen NGC pytorch
25.12-py3, GPU reservada, bind mounts a ai-projects vía NFS y a
~/ft-models en disco local rápido de spark) sin tocar jupyter-pyt
ni el vllm de producción.

docker-compose.yml

@@ -0,0 +1,18 @@
+services:
+  qwen-lora-train:
+    image: nvcr.io/nvidia/pytorch:25.12-py3
+    container_name: qwen-lora-train
+    restart: unless-stopped
+    command: ["sleep", "infinity"]
+    working_dir: /workspace/ai-projects/qwen3-6-lora
+    shm_size: "16gb"
+    volumes:
+      - /mnt/docker-nas/projects/ai-projects:/workspace/ai-projects
+      - /home/aleleba/ft-models:/workspace/ft-models
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]