[Iceberg] Daft Dataframe Reader

[pdames]

To help improve the end-to-end efficiency of compaction and similar operations (e.g. hash-equality joins, dataset filters, hash partitioning, etc.) and based on current benchmarks showing Daft I/O to be more performance that PyArrow and S3FS for S3 Parquet file reads, we would like to add an Daft-native Dataframe reader for Iceberg, where "Daft-native" refers to the desired end state of the implementation relying on no intermediate conversion through a Ray Dataset or any other intermediate format.