Review CALCITE-1054 (operator null semantics)

[Julian Hyde <jh...@apache.org>]

Can someone please review
https://issues.apache.org/jira/browse/CALCITE-1054? I've made a major
change for how we handle null semantics for operators implemented in
Enumerable convention. We now do it by transforming the RexNode tree,
rather than by carrying around context during code generation about
what expressions are known to be not-null.

The effect is that every Java expression that is known to be NOT NULL
can can be stored in a primitive type will be stored in a primitive
type. So an INTEGER NOT NULL (and for that matter a DATE NOT NULL)
will be stored in a primitive int variable.

If x is nullable and we evaluate 'WHERE x > 10 AND x < 20' we first
check whether x is null, and if so return false, and thereafter use x
as a primitive int.

Julian