Postgres function slower than same ad hoc query - postgresql

Question

I have had several cases where a Postgres function that returns a table result from a query is much slower than running the actual query. Why is that?
This is one example, but I've found that function is slower than just the query in many cases.
create function trending_names(date_start timestamp with time zone, date_end timestamp with time zone, gender_filter character, country_filter text)
returns TABLE(name_id integer, gender character, country text, score bigint, rank bigint)
language sql
as
$$
select u.name_id,
n.gender,
u.country,
count(u.rank) as score,
row_number() over (order by count(u.rank) desc) as rank
from babynames.user_scores u
inner join babynames.names n on u.name_id = n.id
where u.created_at between date_start and date_end
and u.rank > 0
and n.gender = gender_filter
and u.country = country_filter
group by u.name_id, n.gender, u.country
$$;
This is the query plan for a select from the function:
Function Scan on trending_names (cost=0.25..10.25 rows=1000 width=84) (actual time=1118.673..1118.861 rows=2238 loops=1)
Buffers: shared hit=216509 read=29837
Planning Time: 0.078 ms
Execution Time: 1119.083 ms
Query plan from just running the query. This takes less than half the time.
WindowAgg (cost=44834.98..45593.32 rows=43334 width=25) (actual time=383.387..385.223 rows=2238 loops=1)
Planning Time: 2.512 ms
Execution Time: 387.403 ms
Buffers: shared hit=100446 read=50220
-> Sort (cost=44834.98..44943.31 rows=43334 width=17) (actual time=383.375..383.546 rows=2238 loops=1)
Sort Method: quicksort Memory: 271kB
Sort Key: (count(u.rank)) DESC
Buffers: shared hit=100446 read=50220
-> HashAggregate (cost=41064.22..41497.56 rows=43334 width=17) (actual time=381.088..381.906 rows=2238 loops=1)
" Group Key: u.name_id, u.country, n.gender"
Buffers: shared hit=100446 read=50220
-> Hash Join (cost=5352.15..40630.88 rows=43334 width=13) (actual time=60.710..352.646 rows=36271 loops=1)
Hash Cond: (u.name_id = n.id)
Buffers: shared hit=100446 read=50220
-> Index Scan using user_scores_rank_ix on user_scores u (cost=0.43..35077.55 rows=76796 width=11) (actual time=24.193..287.393 rows=69770 loops=1)
-> Hash (cost=5005.89..5005.89 rows=27667 width=6) (actual time=36.420..36.420 rows=27472 loops=1)
Rows Removed by Filter: 106521
Index Cond: (rank > 0)
Filter: ((created_at >= '2021-01-01 00:00:00+00'::timestamp with time zone) AND (country = 'sv'::text) AND (created_at <= now()))
Buffers: shared hit=99417 read=46856
Buffers: shared hit=1029 read=3364
Buckets: 32768 Batches: 1 Memory Usage: 1330kB
-> Seq Scan on names n (cost=0.00..5005.89 rows=27667 width=6) (actual time=0.022..24.447 rows=27472 loops=1)
Rows Removed by Filter: 21559
Filter: (gender = 'f'::bpchar)
Buffers: shared hit=1029 read=3364
I'm also confused on why it does a Seq scan on names n in the last step since names.id is the primary key and gender is indexed.