Drop/Ignore back-filled points

Nobody has claimed this yet.

We have a deadman alert to fire if we don't see points in some measurement after 5 minutes. It works great in steady state, but we have had some false negatives due to older points getting backfilled into InfluxDB.

The general sequence of events is:
- Points stop being produced. 5 mins later alert fires
- We backfill points from 2 weeks ago. Alert goes back to OK state

We would like to avoid getting into an OK state if a point older than 5 minutes is inserted for that measurement. How can we only consider points with a relevant timestamp?

The tickscript looks like this:

```
var b = stream
|from()
.database('myDb')
.retentionPolicy('autogen')
.measurement('myMeasurement')
.where(lambda: "field" =~ /^regex$/)
b
|deadman(1.0, 5m, lambda: (MATCHES_CRON))
.id('alertId')
.message('{{ .ID }} is {{ .Level }}: {{ index .Fields "emitted" | printf "%.0f" }} points seen in last 5m')
.stateChangesOnly()
.post('http://myalertsystem.com')

```

The entry point is the shown TICKscript stream and deadman pipeline; start by tracing how deadman counts points within its 5m window and handles point timestamps. Done means inserting points older than five minutes cannot move the alert from fired back to OK, while current points still do.

Written by the indexing model from the issue text.