A per-segment cache of documents' facet ordinals. Every
CachedOrds holds the ordinals in a raw {@code int[]}, and therefore consumes as much RAM as the total number of ordinals found in the segment, but saves the CPU cost of decoding ordinals during facet counting.
NOTE: every CachedOrds is limited to 2.1B total ordinals. If that is a limitation for you then consider limiting the segment size to fewer documents, or use an alternative cache which pages through the category ordinals.
NOTE: when using this cache, it is advised to use a DocValuesFormat that does not cache the data in memory, at least for the category lists fields, or otherwise you'll be doing double-caching.
NOTE: create one instance of this and re-use it for all facet implementations (the cache is per-instance, not static).