SQL Storage: don't crash on duplicate names

With high parallelism, it's possible to create two documents with the same low-level name.
This condition cannot be detected by the framework as it would mean checking things at transaction commit time, which is too late.
However we must not crash when looking up by name a document and finding several.