isthmus: a hint's output names are dropped whenever a column is a struct, because nested field names are compared

[nielspardon]

applyOutputNames drops a hint's names unless the projection's columns are the relation's columns type by type, which describesColumnsOf decides:

substrait-java/isthmus/src/main/java/io/substrait/isthmus/SubstraitRelNodeConverter.java

Lines 1264 to 1273 in 934a60e

	private boolean describesColumnsOf(List<Type> fields, RelDataType rowType) {
	for (int field = 0; field < fields.size(); field++) {
	RelDataType declared = typeConverter.toCalcite(typeFactory, fields.get(field));
	if (!SqlTypeUtil.equalSansNullability(
	declared, rowType.getFieldList().get(field).getType())) {
	return false;
	}
	}
	return true;
	}

SqlTypeUtil.equalSansNullability compares nested ROW field names, and typeConverter.toCalcite(typeFactory, fields.get(field)) is called without a name list, so a struct column's declared type carries placeholder nested names while the row type built from the schema carries the real ones. The two never match, and every name is dropped.

A virtual table with the schema [plain i32, outer struct(a i32, b fp64)], emit = [1] and hint names (LABEL, a, b) comes back named [outer]; the same table with a flat second column comes back named [label].

applyOutputNames only ever restates top-level names -- its own Javadoc says so -- so comparing sans field names for nested structs is enough.

Found while reviewing #1189.