planner: put all cardinality estimation code into a single separate package

[qw4990]

Enhancement

Currently, the code of CE is coupled with other modules, and you can find its code in:

1. The stats package: HistColl.Selectivity, HistColl.GetRowCountByXXX, HistColl.crossValidationSelectivity, etc.
2. Physical optimization: fullJoinRowCountHelper.estimate, DataSource.getOriginalPhysicalTableScan, etc.
3. Logical Optimization: baseLogicalPlan.recursiveDeriveStats, LogicalPlan.DeriveStats, etc.

This makes this module hard to maintain and evolve.

An ideal architecture is shown below, where boundaries between the CE module and other modules are clear:

1. At the bottom, the Stats module provides some low-level interfaces to directly estimate cardinality for range or point on those statistical structures like Hist, TopN, etc.
2. Based on Stats, the CE module provides some high-level interfaces to estimate cardinality for CNF, Join, Agg, etc. All estimation strategies should be put in this separate package, e.g.:
   2.1. How to estimate for Join (fullJoinRowCountHelper.estimate)?
   2.2. How to handle ModifyCnt?
   2.3. How to handle out-of-range estimation (outOfRangeEQSelectivity)?
   2.4. How to prioritize index statistics and column statistics?
   2.5. How to use multiple sorts of statistics to make the estimation result more accurate (crossValidationSelectivity)?
   2.6. ...
3. On the top, all other modules use the interfaces provided by CE to do cardinality estimation.

We decided to refactor(reorganize) related code by following the above design.