task runs manually but not scheduled

Hello,

I have community influx 2.6.1, with the following task:
```
import "join"

option task = {name: "cq_cerit_cpu_used_perc_by_cluster", every: 1h, offset: 10m}

data_avail =
from(bucket: "pbs-cerit")
|> range(start: -3h, stop: now())
|> filter(fn: (r) => r["_measurement"] == "av_state")
|> filter(fn: (r) => r["_field"] == "cpu_avail")
|> group(columns: ["cluster"])
|> aggregateWindow(fn: sum, every: 1m)

data_used =
from(bucket: "pbs-cerit")
|> range(start: -3h, stop: now())
|> filter(fn: (r) => r["_measurement"] == "av_state")
|> filter(fn: (r) => r["_field"] == "cpu_used")
|> group(columns: ["cluster"])
|> aggregateWindow(fn: sum, every: 1m)

data_perc =
join.tables(
method: "inner",
left: data_used,
right: data_avail,
on: (l, r) => l._time == r._time,
as: (l, r) => ({l with avail: r._value}),
)
|> filter(fn: (r) => r["avail"] > 0 and r["_value"] >= 0)
|> map(fn: (r) => ({r with _value: r._value * 100.0 / r.avail}))

data_perc
|> aggregateWindow(fn: mean, every: 1h)
|> map(fn: (r) => ({r with _field: "percent", _measurement: "cpu_used_perc_by_cluster"}))
|> to(bucket: "pbs-cerit-agr")
```

it works ok, if I run it manually, but nothing is produced when running schedulled. If I remove the `aggregateWindow(fn: mean, every: 1h)`, it works even schedulled.

Is this a bug?

Start with the supplied task definition and compare its manual run with scheduled execution, keeping aggregateWindow(fn: mean, every: 1h) in the reproduction. Verify whether scheduled runs produce the expected write to pbs-cerit-agr; done when the same task succeeds on schedule.

Written by the indexing model from the issue text.