cannot trim rows to a unqiue tag value and count them in flux

__Steps to reproduce:__
List the minimal actions needed to reproduce the behavior.

1. Run the proctat input plugin with telegraf pointed at influxdb

sample telegraf config

```toml
[agent]
interval="10s"
flush_interval="10s"

[[inputs.procstat]]
exe = ".*"

[[outputs.influxdb_v2]]
## The URLs of the InfluxDB cluster nodes.
##
## Multiple URLs can be specified for a single cluster, only ONE of the
## urls will be written to each interval.
## urls exp: http://127.0.0.1:9999
urls = ["http://localhost:9999"]

## Token for authentication.
token = "$INFLUX_TOKEN"

## Organization is the name of the organization you wish to write to; must exist.
organization = "rg"

## Destination bucket to write into.
bucket = "bucket1"
```

2. enter the following query and click view raw data:

```
processes = from(bucket: "bucket1")
|> range(start: v.timeRangeStart, stop: v.timeRangeStop)
|> filter(fn: (r) =>
(r["_measurement"] == "procstat"))
|> filter(fn: (r) =>
(r["_field"] == "memory_usage"))
|> keep(columns: ["process_name"])
|> unique(column: "process_name")

processes
|> count()
|> yield(name: "num processes")

processes
|> yield(name: "process names")
```

and I get:


3. Update query to make `count` use the `process_name` field:

```
processes = from(bucket: "bucket1")
|> range(start: v.timeRangeStart, stop: v.timeRangeStop)
|> filter(fn: (r) =>
(r["_measurement"] == "procstat"))
|> filter(fn: (r) =>
(r["_field"] == "memory_usage"))
|> keep(columns: ["process_name"])
|> unique(column: "process_name")

processes
|> count(column: "process_name")
|> yield(name: "num processes")

processes
|> yield(name: "process names")
```

and I get:


__Expected behavior:__
Expected to be able to count rows here.

__Actual behavior:__
Errors, as described in the above

__Environment info:__

* System info: `Darwin 18.7.0 x86_64`
* Influxd version: master(b12c3ef77b)
* Browser: FireFox 79.0

Reproduce the issue with the supplied Telegraf procstat configuration and Flux query, starting with the unique and count operations shown in the report. Trace why counting the resulting rows fails, then verify that both count() forms return the expected process count without errors.

Written by the indexing model from the issue text.