But VM doesn't appear to support fan-out federated query or using object storage, so I can't just drop it in to replace Thanos too.

In VM ecosystem fan-out queries aren't needed. Usually, Prometheus (or stateless scrape agents) is used for scraping and delivering metrics to central VM cluster. Data usually has about 30-60s freshness and can be queried right away from central cluster, providing global query view.

Yes, VictoriaMetrics doesn't support object storage for historical data. But it is very efficient in terms of data compressing, so probably storing everything on disks would cost the same money and will provide better query performance.