Flush flatten node when values for all pipelines are received

Nobody has claimed this yet.

Seems like the flatten node is waiting for new data with greater timestamp to flush the present data.

Below is one sample simplified tick script that we use. Data arrives at a frequency of 10 seconds. So in this case, though the flatten node has all the data for time 't1', it waits for the new data 't1 + 10s' to arrive, to flush the 't1' data down the pipeline. There might be valid use cases in which it might be useful to wait in this way.

**Questions/problems that we have:**

- In our case, it is very important that we process the data as and when it arrives without any delays. Is there a way to get around this? Is there a provision to flush the data at flatten node?

- Also since the data is being held in the memory, in scale scenarios Kapacitor uses huge amount of memory. For a single device, on a limited scale scenario, it used approximately 50GB of RAM. Think of same scenario being applied for multiple devices. The memory usage multiplies by the factor of number of devices.

```
var var_in_pkts = stream
|from()
.database('Test')
.retentionPolicy('autogen')
.measurement('raw')
|eval(lambda: int("in-pkts"), lambda: string("index"), lambda: string("name"))
.as('in', 'id', 'name')
| log()
.prefix('in_stream')
var var_out_pkts = stream
|from()
.database('Test')
.retentionPolicy('autogen')
.measurement(''interfaces'')
|eval(lambda: int("out-pkts"), lambda: string("index"), lambda: string("name"))
.as('out', 'id', 'name')
| log()
.prefix('out_stream')
var joined_fields = var_in_pkts| union(var_out_pkts)
|groupBy('name', 'id')
|flatten()
.tolerance(1s)
|default()
.field('in-pkts',-2147483648)
.field('out-pkts',-2147483648)
|eval(lambda: string("id"), lambda: int("in"), lambda: string("name"), lambda: int("out"))
.as('id', 'in', 'name', 'out')
.tags('id', 'name')
|log()
.prefix('joined_fields')
joined_fields
|influxDBOut()
.measurement('processed')
.database('Test')
.flushInterval(100ms)
```

**Below are the setup details:**
OS: Ubuntu 16.04.2
RAM: 128 GB
Kapacitor version: 1.4.1 (Tested scenarios of pre-built package directly on host and on docker)

The issue names the flatten() node and provides a simplified TICK script, but no source files or tests. Start by tracing the flatten node's buffering and timestamp handling, then reproduce the script's delayed flush and memory growth. Done requires an agreed flush condition for complete pipeline values and tests covering that behavior.

Written by the indexing model from the issue text.