The `LoadHolonsByMetaDataAsync` query is powerful but potentially expensive across heterogeneous backends. What's the current behaviour when the active provider doesn't support efficient key-value metadata queries? Is there a scan fallback, and if so, how does it perform at scale?