Port formatters for Parquet files to Gen3, including support for lazy loading of columns.
Convert pipe_analysis tasks to PipelineTasks
Decide on behavior of ParquetTable if one requested column does not exist
I'm pretty confident the behavior here is correct, given the tests I've written. I'm less confident that it's (all of) the behavior we need, and even less confident that the documentation uses the right terminology for Pandas things.
I'm not confident that it's all the behavior we need either, but that will be revealed when I move on to writing the Gen3 versions of the writeObjectTable and transformCatalog tasks this afternoon. I'm OK with your merging this now if you open up a new ticket to address stuff that comes up when using it.