[VL] Rework the workaround for empty schema batch

[zhztheplayer]

Description

Velox backend uses a very specific way including some workarounds to handle empty schema batches. Over time, this part of code gets bloated by one workaround over another. We'd review all the relevant code and do a complete refactor.

Note it's easier to come up with a solution after we removed Spark 3.2 support. See #540. But we don't wait for that and do the essentials in advance.