feat: select a storage provider at runtime

[kauandotnet]

Why

@nestm/storage shipped one provider factory: createS3StorageDriver. Any application that wanted to run on GCS, Azure, R2, MinIO, or a plain filesystem had to import files-sdk/<provider> itself and wrap the adapter through the bridge — which puts provider SDK types back in the consumer, the exact leak this package exists to prevent.

That also means a provider cannot be named at runtime. An application shipping to more than one environment has to hard-code a driver per backend and choose between them with its own switch.

What

@nestm/storage/files-sdk/provider — createProviderStorageDriver({ provider, config, prefix }) takes the slug as data and resolves that provider's adapter, and only that one, through a lazy import. A deployment selects its store with an environment variable and installs a single native SDK.

config is one flat bag of provider settings (bucket/region for an object store, root for the filesystem, accountName/container for Azure); each provider reads what it needs and ignores the rest, so the shape survives a provider change. The caller's own driver options — prefix, hooks, plugins, readonly, retries — still apply, because the client loadFiles builds is discarded and the bridge builds its own over the resolved adapter.

The same entry point exposes the provider catalog as pure data — listStorageProviders, getStorageProvider, listStorageProviderEnvVars, listStorageProviderSecretEnvVars, isStorageProvider — so startup validation, health checks, and config UIs can read a provider's env contract without loading an adapter. An unknown slug fails closed with INVALID_ARGUMENT before anything is imported.

@nestm/storage/files-sdk/fs — createFsStorageDriver, mirroring the S3 factory. The adapter reaches only node:fs, so unlike every object-store provider it adds no native SDK to an install.

@nestm/storage/files-sdk/s3 — the S3 capability set is split out of createS3StorageDriver as withS3Capabilities(base, options), so the provider factory can keep conditional-copy promotion and the signed policies for the s3 slug without re-deriving S3AdapterOptions from flat config. The S3-compatible wrappers keep only what they declare themselves. EnhancedS3Adapter is renamed S3StorageAdapter and exported; it was previously private.

Notes

• No behavior change to createS3StorageDriver — it now composes the extracted helper. Its existing spec passes untouched.
• files-sdk types stay inside the files-sdk bridge subtree, matching how s3/index.ts already re-exports S3AdapterOptions.

Verification

pnpm run check (oxlint, prettier, tsc), pnpm run test — 49 unit + 22 e2e, pnpm run verify:pack — build, publint --strict, packed-consumer check. All green.

New coverage: filesystem round-trip and prefix scoping, provider resolution by name, driver options reaching the resolved adapter, readonly honored for any provider, the s3 slug keeping its capabilities, a provider that declares no conditional copy not claiming one, unknown-slug rejection, and the catalog surface.