fix: day returned by InfluxDB query

The function `observeRetrievalResultCodes()` includes the following
statement in the InfluxDB query fetching `(day, code, rate)` data:

    |> aggregateWindow(every: 1d, fn: mean, createEmpty: false)

Such query produces a list of values like this:

```
2024-11-15T00:00:00Z,CONNECTION_REFUSED,0.0022313570194142725
2024-11-16T00:00:00Z,CONNECTION_REFUSED,0.002153071995819862
(...)
2025-02-12T00:00:00Z,CONNECTION_REFUSED,0.021266890041248942
2025-02-12T13:08:20.817239423Z,CONNECTION_REFUSED,0.02153170594662248
```

Notice there are two rows for today (2025-02-12). One row contains
data from yesterday (full day) and another row contain partial data from
today.

In this commit, I am fixing the query to correctly assign data points
from yesterday to yesterday's date:

      |> aggregateWindow(every: 1d, fn: mean, createEmpty: false, timeSrc: "_start")

The new query produces a list like this:

```
2024-11-14T00:00:00Z,CONNECTION_REFUSED,0.0022313570194142725
2024-11-15T00:00:00Z,CONNECTION_REFUSED,0.002153071995819862
(...)
2025-02-11T00:00:00Z,CONNECTION_REFUSED,0.021266890041248942
2025-02-12T00:00:00Z,CONNECTION_REFUSED,0.02153170594662248
```

This fixed the error introduced by cbb3bf1 (#316), where the SQL query
fails with the following message:

```
ON CONFLICT DO UPDATE command cannot affect row a second time.
Ensure that no rows proposed for insertion within the same command have duplicate constrained values.
```

See also InfluxDB documentation for `aggregateWindow()`:
https://docs.influxdata.com/flux/v0/stdlib/universe/aggregatewindow/#timesrc

Signed-off-by: Miroslav Bajtoš <[email protected]>

observer/lib/observer.js

@@ -113,7 +113,7 @@ export const observeRetrievalResultCodes = async (pgPoolStats, influxQueryApi) =
       |> range(start: 0)
       |> filter(fn: (r) => r["_measurement"] == "retrieval_stats_honest")
       |> filter(fn: (r) => strings.hasPrefix(v: r._field, prefix: "result_rate_"))
-      |> aggregateWindow(every: 1d, fn: mean, createEmpty: false)
+      |> aggregateWindow(every: 1d, fn: mean, createEmpty: false, timeSrc: "_start")
       |> keep(columns: ["_value", "_time", "_field"])
       |> map(fn: (r) => ({ r with _field: strings.replace(v: r._field, t: "result_rate_", u: "", i: 1) }))
   `)