Actually we will need this when handling normalizers on Elasticsearch 5.1 and below: in these versions, we only have a single namespace for both analyzers and normalizers (since normalizers are translated to analyzers), but we still have separate references for analyzers and normalizers.

So we could, in theory, end up with an analyzer defined with name "foo" in the first version of the SearchFactory, and a normalizer with the same name defined in the second version of the SearchFactory, with a separate reference to each of them, thereby creating conflicting analyzer definitions.