Empty `UNWIND` followed by `MERGE` and a null-filtered `WITH` causes a PostgreSQL backend segmentation fault

[YGY-001]

Docker image `apache/age:release_PG18_1.7.0

Reproduction

LOAD 'age';
SET search_path = ag_catalog, public;
SELECT ag_catalog.create_graph('age_issue_case8_reduced');

SELECT *
FROM ag_catalog.cypher('age_issue_case8_reduced', $$
UNWIND [] AS x
MERGE (n)
WITH x, head(range(9, 23)) AS y
WHERE y IS NULL
RETURN {x: x, y: y} AS row
$$) AS (row agtype);

Expected

The empty input produces zero rows. The query completes normally, returns
zero rows, and does not create a vertex or terminate the PostgreSQL backend.

Actual

server closed the connection unexpectedly
client backend ... was terminated by signal 11: Segmentation fault

[gregfelice]

Confirmed still crashing on current master (4d9b2506, AGE 1.8.0, PostgreSQL 18.4).

Backtrace at the fault:

#0  0x00007fc0a258fc25 in create_cypher_merge_path (root=root@entry=0x55a1934c3e18, rel=rel@entry=0x55a1935b6350, custom_private=0x55a1935b22f0) at src/backend/optimizer/cypher_pathnode.c:200
#1  0x00007fc0a258fe78 in handle_cypher_merge_clause (rti=<optimized out>, rte=<optimized out>, root=0x55a1934c3e18, rel=0x55a1935b6350) at src/backend/optimizer/cypher_paths.c:229
#2  set_rel_pathlist (rti=<optimized out>, root=0x55a1934c3e18, rel=0x55a1935b6350, rte=0x55a1935b02f8) at src/backend/optimizer/cypher_paths.c:89
#3  set_rel_pathlist (root=0x55a1934c3e18, rel=0x55a1935b6350, rti=<optimized out>, rte=0x55a1935b02f8) at src/backend/optimizer/cypher_paths.c:64

Note this crash is in the planner, not the executor — create_cypher_merge_path at src/backend/optimizer/cypher_pathnode.c:200, below handle_cypher_merge_clause and set_rel_pathlist. It never reaches execution, which makes it distinct from every other crash in the current backlog.

For context: I triaged the current fuzz-report backlog against master today. There are six distinct crash sites in it, not one — the largest group is five issues (#2501, #2508, #2527, #2530, #2535) that all share a single NULL dereference in create_vertex, written up in #2535. This issue is not part of that group; it is its own defect and needs its own fix.