Support PERCENT_RANK aggregate function

[jtaylor-sfdc]

Support an aggregate function that calculates the percent rank of an expression like this:

SELECT PERCENT_RANK(<perc>) WITHIN GROUP (ORDER BY <expression> ASC)
FROM t

The PERCENT_RANK function would return a type of DECIMAL.

[anoopsjohn]

We need to support PERCENT_RANK() for every row as shown in here [See Analytic Example part] ?
Here the issue is, we don't support having a column in select which is not in the GROUP BY (ERROR 1018 (42Y27): Aggregate may not contain columns not in GROUP BY). Means we don't need support for PERCENT_RANK() to know the rank of every row column?

[jtaylor-sfdc]

We'll need to treat the column reference in the WITHIN GROUP (ORDER BY salary DESC) as an aggregation even though it's not in the GROUP BY clause. To do this, you'll want to:

• Define a new intermediate ParseNode, something like a PartitionedAggregateFunctionParseNode derived from AggregateFunctionParseNode and add an accept method implementation similar to the other ones.
• In ParseNodeVisitor, add a visitEnter(PartitionedAggregateFunctionParseNode node) and a corresponding visitLeave. You'll need to fill in the implementation at various intermediate classes (just follow what's done for others), with the most important in ExpressionCompiler one being described below.
• In ExpressionCompiler, for the implementation of visitEnter, you'd set the this.aggregateFunction to the PartitionedAggregateFunctionParseNode node and then clear it on the visitLeave (if it's set to the node). Then at compile time, Phoenix will treat that entire node as an aggregate expression.

[jtaylor-sfdc]

Closing as fixed, as we now support PERCENTILE_RANK. Additional work is necessary to support the OLAP extensions, but this is covered by #23.