Timeseries collections should always have a clustered index, which is implicitly added in the createCollection path.

However with viewless timeseries collections, collection cloning code (e.g. chunk migration, shard collection) may incorrectly call into listCollections without rawData, then pass the options directly into lower catalog layers, which ends up creating a timeseries collection without a clustered index.

While the inconsistency is normally detected by existing validations, we should add a tripwire assertion around DatabaseImpl::_createCollection to fail as fast as possible.