Query returns unimplemented error when nesting avg(avg_over_time())

[lmickh]

Describe the bug
After upgrading from 2.3 to 2.5, this query will result in an unimplemented error:

avg(avg_over_time({job=~"myapps.*"} |= "stats" | json busy="utilization" | unwrap busy [$__interval]))

The utilization field being extracted is a float (ex: 55.4). The avg_over_time() appears to properly calculate the values for each individual stream, but when wrapped with a avg() to get the average across all of the streams the unimplemented error is returned instead.

To Reproduce
Steps to reproduce the behavior:

1. Started Loki v2.5.0
2. Started Promtail v2.5.0 to tail Kubernetes pods
3. Query via Grafana v8.4.5: avg(avg_over_time{job="somejob" | json somefield="field" | unwrap somefield [$__interval]))

Expected behavior
While wrapping an avg_over_time() with a avg() is probably not as good as avg_over_time() by (label), there is not an obvious reason why it should not work. I would expect an unwrapped float per log stream could be aggregated with avg().

Environment:

• Infrastructure: EKS
• Deployment tool: Raw manifests via ArgoCD

Work-around

Using avg_over_time(...) by (label) to aggregate instead of wrapping it with avg().

This might be related to #6130 and the changes in 2.4, but I haven't had a chance to test that yet.

[owen-d]

Confirmed, this is in main as well as 2.5 (I'm pretty sure it's been around for a long time).