In BF-13897, it appears we've had a case where two index entries had the same name. This should never happen, but could theoretically happen if the in-memory state was out of date with that of the snapshot held by the thread adding the index entry.

Add an fassert or invariant to DurableCatalogImpl::prepareForIndexBuild to catch this at the moment it occurs, rather than much later when the duplicate entries cause a problem.