Postgres graph query gets slow because of a nested loop join

Context

I am using postgres as a multi-tenant graph database.

All data is stored in a single table, and I rely on partial indexes to make queries efficient.

Here's how a minimal schema looks:

CREATE TABLE triples (
    app_id text NOT NULL,
    entity_id text NOT NULL,
    attr_id text NOT NULL,
    value jsonb NOT NULL,
    eav boolean NOT NULL DEFAULT false,
    ave boolean NOT NULL DEFAULT false,
    vae boolean NOT NULL DEFAULT false,
    created_at bigint NOT NULL DEFAULT 0,
    checked_data_type text
);

CREATE OR REPLACE FUNCTION triples_extract_number_value(value jsonb)
RETURNS double precision AS $$
BEGIN
  IF jsonb_typeof(value) = 'number' THEN
    RETURN value::double precision;
  ELSE
    RETURN NULL;
  END IF;
END;
$$ LANGUAGE plpgsql IMMUTABLE;

CREATE INDEX vae_idx ON triples (app_id, value, attr_id, entity_id) WHERE vae;

CREATE INDEX ave_number_idx ON triples (app_id, attr_id, triples_extract_number_value(value), entity_id) WHERE ave AND checked_data_type = 'number';

In an app with conversations, groups, and messages, data could be stored as:

{app_id: 'app_1', entity_id: 'convo_1', attr_id: 'title': "My Conversation"} 
{app_id: 'app_1', entity_id: 'group_1', attr_id: 'title': "My Group"} 
{app_id: 'app_1', entity_id: 'convo_1', attr_id: 'group': "'group_1'", vae: true} // this 'links' convos to groups
{app_id: 'app_1', entity_id: 'msg_1', attr_id: 'time': "123", ave: true, checked_data_type: 'number'} // this indexes our 'time' field  
{app_id: 'app_1', entity_id: 'msg_1', attr_id: 'convo': "'convo_1'", vae: true} // this  links `msg_1` to `convo_1`

Goal

I want to satisfy the following query:

"Give me all convos, which belong to group 'group-1', and have a message who's time is greater than 5."

To do this, I wrote the following query, which does 3 self-joins:

SELECT 
    DISTINCT(match_0_0.entity_id)
FROM 
    triples AS match_0_0
JOIN 
    triples AS match_0_1
    ON match_0_1.app_id = match_0_0.app_id
    AND match_0_1.vae = true
    AND match_0_1.attr_id = 'convo'
    AND match_0_1.value = to_jsonb(match_0_0.entity_id)
JOIN 
    triples AS match_0_2
    ON match_0_2.app_id = match_0_1.app_id
    AND match_0_2.ave = true
    AND match_0_2.attr_id = 'time'
    AND triples_extract_number_value(match_0_2.value) >= 5
    AND match_0_2.checked_data_type = 'number'
    AND match_0_2.entity_id = match_0_1.entity_id
WHERE 
    match_0_0.app_id = 'chat_app'
    AND match_0_0.vae = true
    AND match_0_0.attr_id = 'groups'
    AND match_0_0.value = '"group_1"';

Problem

The problem is, this query takes about 14 seconds to satisfy. Running with EXPLAIN (ANALYZE, BUFFERS), shows:

Nested Loop

Looking at EXPLAIN (ANALYZE, BUFFERS), I noticed that the nested_loop join has lots of buffer hits.

If I try disabling nested loop joins, the query resolves in 40 ms:

SET enable_nestloop TO off;

SELECT 
    DISTINCT(match_0_0.entity_id)
FROM 
    triples AS match_0_0
JOIN 
    triples AS match_0_1
    ON match_0_1.app_id = match_0_0.app_id
    AND match_0_1.vae = true
    AND match_0_1.attr_id = 'convo'
    AND match_0_1.value = to_jsonb(match_0_0.entity_id)
JOIN 
    triples AS match_0_2
    ON match_0_2.app_id = match_0_1.app_id
    AND match_0_2.ave = true
    AND match_0_2.attr_id = 'time'
    AND triples_extract_number_value(match_0_2.value) >= 5
    AND match_0_2.checked_data_type = 'number'
    AND match_0_2.entity_id = match_0_1.entity_id
WHERE 
    match_0_0.app_id = 'chat_app'
    AND match_0_0.vae = true
    AND match_0_0.attr_id = 'groups'
    AND match_0_0.value = '"group_1"';

Question

Is there a way I can hint to postgres, so it chooses a better strategy?

Repro

I set up a repro on DB Fiddle, which shows the slow query:

https://www.db-fiddle.com/f/4jyoMCicNSZpjMt4jFYoz5/15620