Currently, within the catalog cache we are not considering retrying on errors which are in the "retriable" error category. Based on a recent, i.e., BF-39979, we have found that tests with stepdowns and an embedded config server could fail if the primary re-election occurs immediately after registering the term, leading to this uassert hit. 
We should investigate if we have enough test coverage for non-retriable commands triggering catalog cache, such as the second write operation in multi-statement transactions or multi-writes.