convert prices with exchangerates from other table SUM and get AVG

What my idea is:

Server: Ubuntu with latest MariaDB

It's a portfolio algo, the user can buy stocks in a currency like USD but his portfolio can be in EURO, so the values must be converted.

1. I need a exchange rate near the created date, if not use a rate from date before.
2. SUM the amounts_lefts * price converted to currency i want (in this case for easier understand it will be a rate of 1, the target currency will submit over query in this case 2.
3. the result should be divides though the sum of amounts_left to get the correct buyIN aka AVG price.

Ive updated the data to make things more simple for test, like currency of source is now the same as target, so the calculation is 1:1

SQL Fiddle is here: http://sqlfiddle.com/#!9/8c7fc59/11
Results running against my DB with querys 

    https://pastebin.com/pSerHN0j

the first 2 results are what i want, the 3. one is from jmvcollaborator where the AVG BuyIN is correct but the amounts are doubled

SQL Export with more Data 

    https://pastebin.com/d1aVX2wm

Updated Question:

The solution over Subquery is working, but this Method is not the best in performance, so im currently looking for a solution without this.

After my Information partitions over should be here a solution, but this is not my kind of knowledge, so maybe anyone can help me here.

I will write some findings here, first i would like to double check that your very first query works, please confirm:

SELECT SUM(amounts_left*price)/SUM(amounts_left) as eBuyIN
FROM transactions as t 
WHERE 1 GROUP BY currency_id,broker_id,portfolio_id,instrument_id; 

If it does not work then let me share some findings with regards of the query we got working which is:

SELECT 
SUM(price)/SUM(mleft) 
FROM(
SELECT
`instrument_id`,
amounts_left AS mleft, 
amounts_left * price * (
SELECT `rate` 
FROM `currency_exchanges` 
WHERE `from_currency_id` = t.currency_id 
AND `to_currency_id` = 2 
AND `date` <= t.created 
ORDER BY `date` DESC LIMIT 1
) AS price
FROM `transactions` AS `t` 
WHERE `action` = 1 
AND `portfolio_id` = 128 
) a

The query cost is 2.72 since it perform a full table scan due the nested query. I took the freedom to try a different approach using CTE and OVER PARTITION BY and the query cost is 0.35 a non unique key look up due the columns inside the partition thus much faster that the previous query, here is the CTE code:

WITH 
cte AS (
SELECT ROW_NUMBER() OVER(PARTITION BY t.currency_id,t.broker_id,t.portfolio_id,t.instrument_id ORDER BY amounts_left DESC
) AS rowNumber,        
SUM(amounts_left) OVER(PARTITION BY t.currency_id,t.broker_id,t.portfolio_id,t.instrument_id
ORDER BY amounts_left DESC ROWS BETWEEN CURRENT ROW AND 1 FOLLOWING) AS amountsleftT,       
SUM(amounts_left*price) OVER(PARTITION BY t.currency_id,t.broker_id,t.portfolio_id,t.instrument_id
ORDER BY amounts_left DESC ROWS BETWEEN CURRENT ROW AND 1 FOLLOWING) AS priceT
FROM currency_exchanges AS ce
JOIN transactions AS t
ON from_currency_id = t.currency_id 
AND to_currency_id = 2 
AND ce.date <= t.created
AND action = 1 and portfolio_id = 128 
)
Select priceT/amountsleftT as Total from cte where rowNumber = 1;

I would suggest to add more data so many different cases can be tested you can keep working on the fiddle and share the link.

Hopefully i have helped somehow.

UPDATE query with Over Partition By was updated cause the aggregates were wrong. IMPORTANT: still the performance is better than the other approaches.
Here is the working query:

WITH cte AS
(
select 
instrument_id,
ROW_NUMBER() OVER(PARTITION BY t.currency_id,t.broker_id,t.portfolio_id,t.instrument_id) AS rowNumber,
Sum(amounts_left* price *ce.rate) OVER(PARTITION BY t.currency_id,t.broker_id,t.portfolio_id,t.instrument_id) AS priceSum,
Sum(amounts_left) OVER(PARTITION BY t.currency_id,t.broker_id,t.portfolio_id,t.instrument_id) amountsleftSum 
from transactions t left join
(select ce.*, row_number() over (partition by from_currency_id order by date desc) as seqnum
from currency_exchanges ce
where
to_currency_id = 2     
order by ce.date desc
) ce
on ce.from_currency_id = t.currency_id and seqnum = 1
WHERE 
ce.date <= t.created 
AND action = 1 AND amounts_left > 0 AND portfolio_id = 128 
)    
SELECT     
priceSum/amountsleftSum as   eBuyIN,
amountsleftSum AS amounts_left,
instrument_id       
FROM cte
WHERE  rownumber = 1;