[Fleet] Visualisations have "Field no longer exists in data view" errors when no data has been ingested

[hop-dev]

Kibana version: 7.11 and above

Describe the bug:

When an integration hasn't ingested any data (or only a subset of data) dashboards show "Field no longer exists in data view"
errors.

This is because since 7.11, data views dynamically generate their fields at runtime. Fleet manually builds the list of fields on the index-pattern saved object, however these are not referenced by kibana when checking fields for the visualisation.

Steps to reproduce:
On a fresh kibana install:

1. Setup fleet, including adding an agent with system monitoring
2. Add an integration which will not collect any data to the agent's policy (e.g apache in this example)
3. View the [Logs Apache] Access and error logs dashboard
4. notice the errors as seen in the screenshot above.
5. (optional) Notice we do have the field defined in the given index patter nby performing the following dev tools request:

GET /.kibana/_search
{
  "query": {
    "term": {
      "type": {
        "value": "index-pattern"
      }
    }
  }
}

Expected behavior:
Something telling the user that there is no data, rather than missing field error.

[elasticmachine]

Pinging @elastic/fleet (Team:Fleet)

[jen-huang]

Thanks for filing this Mark. It is related to the issue I filed here: #99243

I suspect the difference in error messages between this issue and that one is due to each type of visualization handling errors in their own way, some present as more severe looking, and others less severe. I'm not sure what the best path forward is for that.

[jen-huang]

This no longer appears to be an issue in 8.7 after various dashboard changes, this is after running the previous testing steps for Apache:

In some cases dashboards may still have field errors, but they are presented differently: