`avg(toFloat(...))` followed by `UNION` fails with `UNION types double precision and agtype cannot be matched`

[YGY-001]

Reproduction

LOAD 'age';
SET search_path = ag_catalog, public;
SELECT ag_catalog.create_graph('age_issue_case19');

SELECT *
FROM ag_catalog.cypher('age_issue_case19', $$
RETURN avg(toFloat(1)) AS value
UNION ALL
RETURN null AS value
$$) AS (value agtype);

Expected

The query returns two rows: 1.0 and null. The equivalent query is
accepted by Neo4j, and AGE can return the aggregate by itself.

Actual

ERROR: UNION types double precision and agtype cannot be matched

Controls

These queries succeed on the same fresh graph:

RETURN avg(toFloat(1)) AS value

RETURN toFloat(1) AS value
UNION ALL
RETURN null AS value

The failure is specific to combining the numeric aggregate result with a
UNION branch. It does not require a graph pattern, writes, an entity, or a
probe projection. Current-log Case 3197 and its full fixture reproduce this
error on AGE 1.8.0.

This is separate from #2525: #2525 requires a numeric aggregate to be used in
a CASE WHEN ... IS NULL THEN ... ELSE ... END expression and reports a
boolean conversion error. This reduction contains no CASE and fails during
UNION type resolution.

Environment

Apache AGE 1.8.0 on PostgreSQL 18.6, Docker image
apache/age:release_PG18_1.8.0.