incorrect druid SQL results using GROUP BY and COUNT

Some druid SQL queries including GROUP BY and COUNT return incorrect results. I'm including an example with the wikipedia dataset.

### Affected Version

Seen in 0.20.0 and 0.21.0.

### Description

Particular GROUP BY queries are returning 0 counts when they should return non-zero numbers. Providing two examples, one with wikipedia, and one with a custom data set (attached).

With wikipedia, we can see the following, to zero in on the issue:

This query returns various non-zero numbers for the counts:

```
select comment, count(*) FROM wikipedia
WHERE channel = '#ar.wikipedia'
GROUP BY comment
```

This query will return 1's, which makes sense:

```
select comment, count(distinct(channel)) FROM wikipedia
WHERE channel = '#ar.wikipedia'
GROUP BY comment
```

So, we have data for the channel field (which makes sense, since we filter for a channel), but the following query returns 0's for all the counts, which is not only incorrect, but logically impossible given the query:

```
select comment, count(channel) FROM wikipedia
WHERE channel = '#ar.wikipedia'
GROUP BY comment
```

Reproduce the three GROUP BY queries against the Wikipedia dataset on the affected Druid versions, 0.20.0 and 0.21.0, and compare COUNT(*), COUNT(DISTINCT channel), and COUNT(channel). Trace the SQL aggregation path responsible for the results; done means COUNT(channel) returns the expected non-zero counts for the filtered rows.

Written by the indexing model from the issue text.