groupBy not grouping by a tag created with eval

Nobody has claimed this yet.

- Full details of your operating system (or distribution) e.g. 64-bit Ubuntu 14.04.
- Running InfluxDB/Kapacitor/Chronograf as Docker containers on MacOSX, latest Docker.
- The version of Kapacitor you are running
- Latest, 1.4.
- Whether you installed it using a pre-built package, or built it from source.
- Official Docker container

We are running into an issue with TICKscript and its groupBy behaviour.

We have two sets of measurements, indoor_temperatures and outdoor_temperatures, which we query with a batch.

The queries look as follows:

```js
var out_temp = batch
|query('SELECT mean(temperature) FROM yyyy')
.every(10s)
.period(120d)
.groupBy(time(1h))
.fill(0)

var in_temp = batch
|query('SELECT mean(temperature) FROM xxxx')
.every(10s)
.period(120d)
.groupBy(time(1h))
.fill(0)
```

If we HTTP out both of them, they create the following sets of data:

``` json
{
"series": [
{
"name": "outdoor_temperatures",
"columns": [
"time",
"mean"
],
"values": [
[
"2017-09-20T17:00:00Z",
0
],
[
"2017-09-20T18:00:00Z",
11.5
]
... the rest
]
}
]
}

{
"series": [
{
"name": "indoor_measurements",
"columns": [
"time",
"mean"
],
"values": [
[
"2017-09-20T17:00:00Z",
585.44012944984
],
[
"2017-09-20T18:00:00Z",
592.94890510949
]
... the rest
]
}
]
}
```

Now we do a full join of them, which gives us expected results

```js
out_temp
|join(in_temp)
.as('out_temp_mean', 'in_temp_mean')
.tolerance(5m)
.fill(0)
```

httpOut:

```json
{
"series": [
{
"name": "outdoor_temperatures",
"columns": [
"time",
"in_temp_mean.mean",
"out_temp_mean.mean"
],
"values": [
[
"2017-09-20T17:00:00Z",
586.10175438596,
0
],
[
"2017-09-20T18:00:00Z",
592.94890510949,
11.5
]
... the rest
]
}
]
}
```

Which looks perfect. The issue raises when we want to *round the `out_temp_mean.mean` down and groupBy it*

So we go ahead and extend the script

```
out_temp
|join(in_temp)
.as('out_temp_mean', 'in_temp_mean')
.tolerance(5m)
.fill(0)
|eval(lambda: string(floor("out_temp_mean.mean")))
.as('bucket')
.tags('bucket')
.keep('out_temp_mean.mean', 'in_temp_mean.mean')
```

After which the output STILL looks as it should:

``` json
{
"series": [
{
"name": "outdoor_temperatures",
"columns": [
"time",
"in_temp_mean.mean",
"out_temp_mean.mean",
"bucket"
],
"values": [
[
"2017-09-20T17:00:00Z",
586.99190283401,
0,
"0"
],
[
"2017-09-20T18:00:00Z",
592.94890510949,
11.5,
"11"
]
]
}
]
}
```

Now only thing left is to group the values by the new tag bucket:

```js
out_temp
|join(in_temp)
.as('out_temp_mean', 'in_temp_mean')
.tolerance(5m)
.fill(0)
|eval(lambda: string(floor("out_temp_mean.mean")))
.as('bucket')
.tags('bucket')
.keep('out_temp_mean.mean', 'in_temp_mean.mean')
|groupBy('bucket')
```

After which everything goes awry and we are greeted with `series: null`

```
{
"series": null
}
```

Is this expected behaviour? A bug? Or something else?

Reproduce the supplied batch, join, eval, tags, and groupBy TICKscript, comparing the output before and after the final groupBy('bucket'). Trace Kapacitor's handling of tags created by eval and grouped afterward. Done means the behavior is explained and, if confirmed as a bug, the grouped result is returned consistently with the reported bucket values.

Written by the indexing model from the issue text.