Clarify histogram NDV metadata and add real NDV-based estimates

[KKould]

Problem

HistogramMeta::number_of_distinct_value currently looks like NDV, but it is assigned from values_len during analyze. In practice it is used as a minimum selectivity denominator in histogram range estimation, not as true number-of-distinct-values metadata.

This naming can be misleading and may cause future optimizer rules to treat it as real NDV.

Possible Fix

• Rename the current field to reflect its actual meaning, such as values_len, non_null_count, or selectivity_denominator.
• Add separate real NDV metadata later, likely from analyze using HLL or an exact-small/fuzzy-large strategy.
• Use real NDV for optimizer estimates where appropriate:
  • equality selectivity fallback: rows / ndv
  • group by / distinct cardinality
  • join cardinality
  • distinct/group cost
  • index scan equality-prefix fallback cost

Notes

CMS should still be preferred for concrete value frequency when available. Histogram should still drive range estimates. NDV should fill the broader cardinality/selectivity gaps.