Remove the MinIO/S3 storage backend entirely. The filesystem backend is
fully implemented, already used in production, and a migrate-storage tool
exists for any remaining MinIO deployments to migrate beforehand.
Changes:
- Delete MinIO client implementation (internal/storage/storage.go)
- Delete migrate-storage tool (cmd/migrate-storage, scripts/migrate-storage.sh)
- Remove MinIO service, volumes, and env vars from all Docker Compose files
- Simplify StorageConfig: remove Endpoint, AccessKey, SecretKey, Bucket,
UseSSL, Region fields; add SILO_STORAGE_ROOT_DIR env override
- Change all SQL COALESCE defaults from 'minio' to 'filesystem'
- Add migration 020 to update column defaults to 'filesystem'
- Remove minio-go/v7 dependency (go mod tidy)
- Update all config examples, setup scripts, docs, and tests
Extract a FileStore interface from the concrete *storage.Storage MinIO
wrapper so the API layer is storage-backend agnostic.
- Define FileStore interface in internal/storage/interface.go
- Add Exists method to MinIO Storage (via StatObject)
- Add compile-time interface satisfaction check
- Change Server.storage and ServerState.storage to FileStore interface
- Update NewServer and NewServerState signatures
- Add Backend and FilesystemConfig fields to StorageConfig
- Add backend selection switch in main.go (minio/filesystem/unknown)
- Update config.example.yaml with backend field
The nil-interface pattern is preserved: when storage is unconfigured,
store remains a true nil FileStore (not a typed nil pointer), so all
existing if s.storage == nil checks continue to work correctly.
Closes#126
