[Bug]: Chunks are not excluded for a hypertable on the parameterized side of a nested loop

=============== A: join parameter only ===============
 Aggregate (actual rows=1 loops=1)
   ->  Nested Loop (actual rows=72 loops=1)
         ->  Values Scan on "*VALUES*" (actual rows=3 loops=1)
         ->  Append (actual rows=24 loops=3)
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_2_50_chunk m_1 (actual rows=8 loops=3)
                     Index Cond: ((device_id = "*VALUES*".column1) AND (ts >= '2025-02-01 00:00:00+00') AND (ts < '2025-02-02 00:00:00+00'))
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_2_54_chunk m_2 (actual rows=8 loops=3)
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_2_58_chunk m_3 (actual rows=8 loops=3)
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_2_62_chunk m_4 (actual rows=0 loops=3)
               ...  (12 chunk nodes in total, 9 of them with actual rows=0 loops=3)

=============== B: same query + AND ts > now() - interval '100 years' ===============
 Aggregate (actual rows=1 loops=1)
   ->  Nested Loop (actual rows=72 loops=1)
         ->  Values Scan on "*VALUES*" (actual rows=3 loops=1)
         ->  Custom Scan (ChunkAppend) on metric m (actual rows=24 loops=3)
               Chunks excluded during startup: 0
               Chunks excluded during runtime: 0
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_1_2_chunk m_1 (actual rows=8 loops=3)
                     Index Cond: ((device_id = "*VALUES*".column1) AND (ts >= ...) AND (ts < ...) AND (ts > (now() - '100 years'::interval)))
               ...  (12 chunk nodes in total, 9 of them with actual rows=0 loops=3)

=============== C: same keys as plan-time constants ===============
 Finalize Aggregate (actual rows=1 loops=1)
   ->  Append (actual rows=3 loops=1)
         ->  Partial Aggregate (actual rows=1 loops=1)
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_1_2_chunk m_1 (actual rows=24 loops=1)
                     Index Cond: ((device_id = ANY ('{1,2,3}'::bigint[])) AND (ts >= ...) AND (ts < ...))
         ->  Partial Aggregate (actual rows=1 loops=1)
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_1_6_chunk m_2 (actual rows=24 loops=1)
         ->  Partial Aggregate (actual rows=1 loops=1)
               ->  Index Only Scan using ..._chunk_metric_device_id_ts_idx on _hyper_1_10_chunk m_3 (actual rows=24 loops=1)
         (3 chunk nodes in total)

=============== hash partitions touched by keys 1,2,3 ===============
 total_partitions | partitions_touched
------------------+--------------------
               12 |                  3

docker run -d --name tsdb -e POSTGRES_PASSWORD=postgres -e POSTGRES_DB=repro \
  -p 5432:5432 timescale/timescaledb:2.29.1-pg17
psql -h localhost -U postgres -d repro -f repro.sql

CREATE EXTENSION IF NOT EXISTS timescaledb;

CREATE TABLE metric (
    ts        timestamptz NOT NULL,
    device_id bigint      NOT NULL,
    val       double precision
);
SELECT create_hypertable('metric', by_range('ts', INTERVAL '30 days'));
SELECT add_dimension('metric', by_hash('device_id', 12));

INSERT INTO metric
SELECT g, d, random()
FROM generate_series('2025-01-01'::timestamptz, '2025-03-31', '1 hour') g,
     generate_series(1, 500) d;

CREATE INDEX ON metric (device_id, ts);
VACUUM ANALYZE metric;

-- put the hypertable on the parameterized side of a nested loop
SET enable_hashjoin = off;
SET enable_mergejoin = off;
SET enable_material = off;
SET max_parallel_workers_per_gather = 0;

-- A: key comes from a join parameter -> plain Append, all 12 space chunks probed per outer row
EXPLAIN (ANALYZE, COSTS OFF, TIMING OFF, BUFFERS OFF, SUMMARY OFF)
SELECT count(*)
FROM (VALUES (1::bigint), (2), (3)) v(device_id),
LATERAL (
    SELECT 1 FROM metric m
    WHERE m.device_id = v.device_id
      AND m.ts >= '2025-02-01'::timestamptz
      AND m.ts <  '2025-02-02'::timestamptz
) x;

-- B: same query + always-true qual -> ChunkAppend, but "Chunks excluded during runtime: 0"
EXPLAIN (ANALYZE, COSTS OFF, TIMING OFF, BUFFERS OFF, SUMMARY OFF)
SELECT count(*)
FROM (VALUES (1::bigint), (2), (3)) v(device_id),
LATERAL (
    SELECT 1 FROM metric m
    WHERE m.device_id = v.device_id
      AND m.ts >= '2025-02-01'::timestamptz
      AND m.ts <  '2025-02-02'::timestamptz
      AND m.ts > now() - interval '100 years'
) x;

-- C: reference, same keys as plan-time constants -> only 3 chunks in the plan
EXPLAIN (ANALYZE, COSTS OFF, TIMING OFF, BUFFERS OFF, SUMMARY OFF)
SELECT count(*) FROM metric m
WHERE m.device_id = ANY (ARRAY[1, 2, 3]::bigint[])
  AND m.ts >= '2025-02-01'::timestamptz
  AND m.ts <  '2025-02-02'::timestamptz;

-- keys 1,2,3 hash into 3 of the 12 partitions
WITH ids(device_id) AS (SELECT unnest(ARRAY[1, 2, 3]::bigint[])),
slices AS (
  SELECT DISTINCT ds.range_start, ds.range_end
  FROM _timescaledb_catalog.dimension d
  JOIN _timescaledb_catalog.hypertable h ON h.id = d.hypertable_id AND h.table_name = 'metric'
  JOIN _timescaledb_catalog.dimension_slice ds ON ds.dimension_id = d.id
  WHERE d.column_name = 'device_id'
)
SELECT (SELECT count(*) FROM slices) AS total_partitions,
       count(DISTINCT s.range_start) AS partitions_touched
FROM ids
JOIN slices s ON _timescaledb_functions.get_partition_hash(ids.device_id) >= s.range_start
             AND _timescaledb_functions.get_partition_hash(ids.device_id) <  s.range_end;