Index for threadId on memory_messages vector table

WITH vector_scores AS (
  SELECT
    vector_id as id,
    1 - (embedding <=> '[0.01833203,(...),0.038586926]'::vector) as score,
    metadata
  FROM "memory_messages"
  WHERE
    CASE
      WHEN 'xxx-some-thread-id'::text IS NULL THEN metadata#>>'{thread_id}' IS NULL
      ELSE metadata#>>'{thread_id}' = 'xxx-some-thread-id'::text
    END
)
SELECT
  *
FROM vector_scores
WHERE
  score > -1
ORDER BY
  score DESC
LIMIT 3

Limit  (cost=1102519.02..1102519.37 rows=3 width=153)
  ->  Gather Merge  (cost=1102519.02..1107418.90 rows=41996 width=153)
        Workers Planned: 2
        ->  Sort  (cost=1101519.00..1101571.49 rows=20998 width=153)
              Sort Key: (('1'::double precision - (memory_messages.embedding <=> '[0.01833203,(...),0.038586926]'::vector))) DESC
              ->  Parallel Seq Scan on memory_messages  (cost=0.00..1101247.60 rows=20998 width=153)
                    Filter: (((metadata #>> '{thread_id}'::text[]) = 'xxx-some-thread-id'::text) AND (('1'::double precision - (embedding <=> '[0.01833203,(...),0.038586926]'::vector)) > '-1'::double precision))

CREATE INDEX CONCURRENTLY idx_memory_messages_thread_id 
ON memory_messages ((metadata->>'thread_id'));