Major bugfix for multiple annotations in API querysets

[SchrodingersGat]

Problem

When performing multiple aggregation annotations on a queryset, the returned results are sometimes not correct. There is a very long discussion here

I thought I had fixed this in a previous PR by adding distinct=True to the annotations. However, this was not the correct solution. The distinct operator only aggregates values which are distinct (unique) and while this resulted in correct behavior in my test cases at the time, it was not sufficient.

Reproduce

This problem only appears in the REST API which uses queryset annotations to improve database performance.

To reproduce, create multiple stock items for a Part with the same quantity. For example, a Part has 10x stock items, each with quantity 5.

Instead of reporting quantity=50 the API data will report quantity=5 (because it ignores rows which do not have "unique" quantity values)...

Solution

The "correct" solution appears to be to use the Django Subquery function.

However this is really unwieldy and difficult to understand.

Luckily, the django-sql-utils module makes this very easy!

Aggregating stock quantity against parts is now as simple as:

        # Filter to limit stock items to "available"
        stock_filter = Q(
            status__in=StockStatus.AVAILABLE_CODES,
            sales_order=None,
            build_order=None,
            belongs_to=None,
            customer=None,
        )

        # Annotate with the total 'in stock' quantity
        queryset = queryset.annotate(
            in_stock=Coalesce(
                SubquerySum('stock_items__quantity', filter=stock_filter),
                Decimal(0)
            ),
        )

Fixes #954