Kryo

Spark recommends [Kryo serialization](https://github.com/EsotericSoftware/kryo).
This issue tracks tasks to ensure that this serialization framework is used and works correctly.
- find and fix cases where generated RDD are not using Kryo (e.g. MEMORY_ONLY)
- [implement](https://github.com/EsotericSoftware/kryo#implementing-a-serializer) built-in [Kryo serialization](https://stackoverflow.com/questions/36144618/spark-kryo-register-a-custom-serializer) to take advantage of the bitfields information that stores which fields are populated
- write and perform benchmark tests to check if it's really better

Start by locating the generated RDD paths and the MEMORY_ONLY cases mentioned in the issue. Then read the linked Kryo serialization guidance and identify how benchmark tests are organized. Done means the listed RDD cases use Kryo, built-in serialization handles the populated-field bitfields, and benchmark results show whether it improves performance.

Written by the indexing model from the issue text.