e2e: flakey TestMetricsLinux test

The [Linux alloc metrics e2e test](https://github.com/hashicorp/nomad/blob/5fbcf82ceefe5facacaff31f45e04ccd89fed3ac/e2e/metrics/metrics.go#L89) occasinally fails with the following error:

```
prometheus query failed (nomad_client_allocs_memory_usage{exported_job="memstress"}): expected non-zero metrics, got: nomad_client_allocs_memory_usage{alloc_id="70ed194c-e5e5-099b-d000-17cc0ad6920e", exported_job="memstress", instance="172.31.13.159:4646", job="nomad_metrics", namespace="default", task="memstress", task_group="memstress"} => 0 @[1606903456.227]
```

Start in e2e/metrics/metrics.go at line 89 and run the TestMetricsLinux end-to-end test to reproduce the intermittent zero-value metric. Trace the Prometheus query for nomad_client_allocs_memory_usage and make the test reliable; done means the test passes consistently without the reported flaky failure.

Written by the indexing model from the issue text.