Pb/iceberg partitioning aware cardinality estimation

[Tishj]

This PR supersedes #778

I've made several changes to the branch, to reduce code duplication and make use of the data that's already present.

I've had to leave part of the fix commented out however, as we need global_columns and global_column_ids to call ProcessDeletes
To reduce future merge conflicts I've left that untouched for now, but likely we have to split that logic into a part that can be done without those, to then follow it up with a method that finalizes the ProcessDeletes logic (which looks to be purely equality-delete focused).

[Tishj]

		// if the above code didn't find an equality filter (i.e country_code = "[us]")
		// and there are other table filters (i.e cost > 50), use default selectivity.
		bool has_equality_filter = (cardinality_after_filters != base_table_cardinality);
		if (!has_equality_filter && has_non_optional_filters) {
			cardinality_after_filters = MaxValue<idx_t>(
			    LossyNumericCast<idx_t>(double(base_table_cardinality) * RelationStatisticsHelper::DEFAULT_SELECTIVITY),
			    1U);
		}

This is causing the estimated cardinality to be reduced from 500 to 100, get_partition_stats doesn't even get hit 😬

[Tishj]

Actually, I think this should be using this instead of reading all the manifest files/data files