Currently we only store double valued deequ metrics.

To report DQ on high cardinality datasets we would like to summarize stats using percentiles, or other distributions.

We should extend the export SerDes, and iceberg tables, to support complex types.