bug: prefilter cannot see multiple row versions

[silver-ymz]

Reproduce Case:

CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3), visible boolean default true);
INSERT INTO items (embedding) SELECT ARRAY[random(), random(), random()]::real[] FROM generate_series(1, 1000);
CREATE INDEX ON items USING vchordrq (embedding vector_l2_ops);
SET vchordrq.prefilter = on;
SET enable_seqscan = off;

SELECT * FROM items WHERE visible = true ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
-- if the nearest id = 10
UPDATE items SET visible = false WHERE id = 10;
UPDATE items SET visible = true WHERE id = 10;

SELECT * FROM items WHERE visible = true ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
-- now we can't get the tuple which id is 10

https://github.com/tensorchord/VectorChord/blob/54dc7b9fd36d73171ff87f1cbf6022a6b1865025/src/index/fetcher.rs#L97-L113
When fetching tuple from heap, we should use index_fetch_tuple rather than tuple_fetch_row_version. The latter does not check MVCC

[usamoi]

fixed in 0.5.2