Implicit window aggregation for difference

[Seddryck]

70 is a prerequisite.

Following the same idea than for #68, it should be possible to easily define a query where we make a transformation of the time series to represent the delta with the previous element.

SELECT
    Instant,
    Producer,
    AGGREGATE(Produced)
FROM
    WindEnergy
WHERE
        Instant < '2023-01-01'
BUCKET BY 
    DAY
GROUP BY
    Producer

is returning the following table

Instant	Producer	Produced
2022-12-28	Green Power Inc.	30
2023-12-29	Green Power Inc.	24.491114
2023-12-30	Green Power Inc.	30
2023-12-31	Green Power Inc.	29.702520
2022-12-28	Future Energy	14.816336
2023-12-29	Future Energy	17.009873
2023-12-30	Future Energy	17.833747
2023-12-31	Future Energy	16.787512

Then the following SQL

SELECT
    Instant,
    Producer,
    VARIATION(Produced)
FROM
    WindEnergy
WHERE
         Instant < '2023-01-01'
BUCKET BY 
    DAY
GROUP BY
    Producer

Instant	Producer	Produced
2022-12-28	Green Power Inc.	NULL
2023-12-29	Green Power Inc.	-4.185732
2023-12-30	Green Power Inc.	4.185732
2023-12-31	Green Power Inc.	0
2022-12-28	Future Energy	NULL
2023-12-29	Future Energy	-0.693749
2023-12-30	Future Energy	6.936893
2023-12-31	Future Energy	-5.635899

The query translated to the dialect supported by DuckDB should be

SELECT
    DAY(Instant),
    Producer, 
    MAX(Produced) - LAG(MAX(Produced)) OVER (PARTITION BY Producer ORDER BY DAY(Instant) ASC) as Producedx 
FROM
    WindEnergy
WHERE
    Instant < '2023-01-01'
GROUP BY 
    DAY(Instant), producer

In the CTE, the aggregation MAX is the default aggregation of the measurement 'Produced'