https://github.com/apache/spark/pull/31858 fixed it to respect Univocity's default buffer to:
- Firstly, it's best to trust their judgement on the default values. Also 128 is too low.
- Default values arguably have more test coverage in Univocity.
- It will also fix uniVocity/univocity-parsers#449 which is a regression compared to Spark 2.4
To prevent related side effect, we should have a workaround to change the buffer size.