Some implementation detail of row count estimation is faulty

[time-and-fate]

Enhancement

1. In (ds *DataSource) DeriveStats, we first call fillIndexPath() then call Selectivity(). This means we first calculate necessary information for every index access path then in Selectivity() we'll use them. However, the information for the primary index access path of a clustered table is calculated in deriveTablePathStats(), which is after calling Selectivity().
   We need to reorder this logic.
2. In crossValidationSelectivity(), we use the lowExclude and the highExclude from idxPointRange. But the idxPointRange may not be the range of a "point". From (coll *HistColl) getIndexRowCount, it can be a "range" (e.g. a multi-column Range where first several columns are "point" and others are "range"). So it's possible that the lowExclude and highExclude are both true, which makes the crossValidationSelectivity() wrongly return 0.
   I think the simplist fix would be setting the lowExclude and the highExclude to false in crossValidationSelectivity().

[time-and-fate]

/sig planner