S3-compatible object storage for uploads

[sebprunier]

Goal

Uploaded files (GeoJSON, CSV, Shapefile, …) are written to a local UPLOAD_DIR. This forces the API and worker to share a filesystem, which is awkward as soon as they run on different hosts or in different containers. Switching to an S3-compatible backend removes that constraint and enables horizontal scaling of workers.

Scope

• A FileStorage interface with two implementations:
  • LocalFileStorage (current behaviour, default)
  • S3FileStorage (any S3-compatible: AWS S3, MinIO, Cloudflare R2, Scaleway, OVH Object Storage)
• Backend selected via STORAGE_BACKEND=local|s3 env var
• S3 config via env: S3_BUCKET, S3_ENDPOINT, S3_REGION, S3_ACCESS_KEY, S3_SECRET_KEY, S3_FORCE_PATH_STYLE
• API stores the file via the FileStorage, queues a job carrying the storage URI; worker reads it back through the same interface

Acceptance criteria

• FileStorage interface and LocalFileStorage refactor (no behavioural change)
• S3FileStorage using @aws-sdk/client-s3 v3
• Selection via STORAGE_BACKEND env var
• Vitest integration tests for the S3 backend, using a Testcontainers MinIO container (no AWS account required)
• Documentation: docs/guide/configuration.md and a deployment example (compose with MinIO)

Out of scope

• Automatic migration of files already on disk (one-time operator task)
• Pre-signed URLs for direct browser uploads (could come later)