aggregateWindow uses wrong _start and _stop values

Without `aggregateWindow`

After applying `aggregateWindow`

Here's a simple exampel of `aggregateWindow` compleetly fucking up the data
```
import "array"
import "date"

d = array.from(rows: [
{_time: 2023-01-01T00:00:00Z, _value: 1},
{_time: 2023-02-01T00:00:00Z, _value: 2},
{_time: 2023-03-01T00:00:00Z, _value: 3}
])

d
|> range(start: 2023-01-01T00:00:00Z, stop: 2023-05-01T00:00:00Z)
|> map(fn: (r) => ({
_start: 2023-04-01T00:00:00Z,
_stop: 2023-04-30T00:00:00Z,
_time: date.add(d: duration(v: string(v: r._value) + "d"), to: 2023-04-10T00:00:00Z),
_value: r._value
}))
|> aggregateWindow(every: 1mo, fn: count)
|> yield()
```

Start by running the supplied Flux query and inspecting the aggregateWindow entry point and its handling of _start and _stop. Trace why the reported window metadata is produced, then add a regression test for this example and confirm the resulting values match the intended window boundaries.

Written by the indexing model from the issue text.