refactor column stats

[time-and-fate]

Development Task

Please refer to Index Selection Work Plan for details about the entire project.

This issue record works on column statistics.

• Introduce stats version field (we're already storing this in mysql.stats_histograms but not used in column stats) to column stats to distinguish old stats and new stats.
• Use a system variable (will be introduced by sessionctx: add variable tidb_analyze_version #21515) to fall back to the old method.
• Switch from EncodeValue to EncodeKey because we need topn to be ordered to handle range query.
• Collect new version stats (no cmsketch + extract topn when building stats from samples)
• Use new version stats (equal cond: topn + bucket.repeats + uniform distribution assumption for others) (range cond: take topn into consideration)
• Correctly save/load new stats to/from storage/JSON.

There also will be further tests.

[time-and-fate]

/assign time-and-fate