Consider Summarize implementation improvements

[MikeStall]

No change to language semantics , but consider some implementation improvements to help with maintainability & delegation.

[1] Comment the implementation

[2] In Summarize, line 1446, we do a uniqueness check:
var key = keyRecord.ToExpression();

This is not ideal. We really shouldn't call ToExpression at runtime like this - I'm wondering if there are cases it could be unpredictable.

This overlaps with Distinct. It would be good to compare how distinct does it.

[3] Consider a "Plan" design pattern.
At the source-level, grouping columns are identifiers.
Aggregate columns are 'As' nodes.

At the IR level, we pass all this through a formulaValue[].
grouping columns are encoded as StringValues
aggregate columns are encoded as LambdaFormulaValue around a LazyEvalNode around a RecordNode , where that Record has 1 field, where the field's Name is the 'As' , and the field's vakue is the lambda to run the expression. (phew, this is starting to get complex!)
This is kind of "hidden" contract between IR and interpreter - it's kind of arbitrary. But Delegation needs to be aware of this.
This is a pretty simple contract, so it's not a big deal. But it's super brittle- there's a log of arbitrary choices in here on encoding, and if we ever changed it, we'd break delegation.

Whereas a "Plan" object would make it an explicit contract. Ie, imagine if the IR captures all of the key information in some structured C# class that it passed through.

Join has a similar thing as is more brittle.

A "plan" object, might look like:

// Eg: Summarize(t1, SalesDate, Sum(ThisGroup.Amount) As Total);
// Binder has guaranteed uniqueness of all column names 
class SummarizePlan
{
	// The columns used to group. Ie "SalesDate". 
       // This is unordered.
	IEnumerable<string> GroupingColumnNames {get;set; }
	// The aggregate columns computed on each group.        
	// This takes in a 'ThisGroup' identifier specified via rule scope. 
       // Key is column name (eg, "Total")
       // Value is the IR to run it,  (eg, Sum(ThisGroup.Amount)) 
	Dict<string, IRNode> AggregateColumns {get;set; }  
 
       // Possibly extra FormulaType info if needed. 
}

[anderson-joyle]

[1] Comment the implementation

Addressed at #2779