Using or not using Pandas

[SemyonSinchenko]

Using pandas and pyarrow may improve the performance of collect operations (like columns_to_list). On the other side, both pandas and pyarrow are optional dependencies for PySpark SQL. Should we use them or not? And if we should, is it a good idea to separate any calls to these libs to allow other functions to work well?

@MrPowers @jeffbrennan