core: a '?'-marked parameter reference keeps the marker in its lookup key, so it can never resolve

[nielspardon]

The grammar's #ParameterName alternative is Identifier isnull=QMark?, so the ? is a separate labelled token. ParseToPojo.visitParameterName builds the POJO from ctx.getText(), which includes the marker, so L? produces StringLiteral{nullable=true, value="L?"} — the name and the marker both end up in value().

Every lookup in TypeExpressionEvaluator.visit(ParameterizedType.StringLiteral) keys off that string: the locals map, boundInteger, parseIntegerLiteral and boundType. None can ever match, so a ?-marked reference is unresolvable, and the two stringLiteral.nullable() reads in that method are unreachable for this node.

Measured on main at cf581f4, with a declared argument varchar<L> bound to varchar(10):

varchar<L>    ->  VarChar{length=10}
varchar<L?>   ->  UnsupportedOperationException: Unbound type parameter 'L?' in return-type expression

visitAnyType gets this right, reading AnyVar().getText() and the isnull token separately, which is why any1? resolves and a named parameter does not.

The fix is .value(ctx.Identifier().getText()). The practically relevant case is a nullable reference to a named type parameter rather than the integer position shown above, which is what makes the dead nullable() handling in the evaluator matter.