[elastic_agent] Have agent status rule templates use alerting_v2 recovery query

## Summary

Replace the large time-window workaround introduced in https://github.com/elastic/integrations/issues/18885 on agent status alerting rule templates with a proper recovery query, once `alerting_v2` exposes one for ES|QL rules.

Implementing PR for recovery queries: https://github.com/elastic/kibana/pull/255876

This issue will depend on the release of alerting_v2.

## Problem

The agent status rule templates are forced to use large time window queries since alerts auto-recover when the trigger query returns zero hits and the `elastic-agent.status_change` captures only status transitions. A stuck-unhealthy agent emits no further docs, so any finite window eventually false-recovers.

`alerting_v2` introduces separate an optional recovery policy query which allows us to define "alert when X happens, recover when Y happens" pattern needed here.

## Acceptance criteria
- Alerts open timely when an agent goes into an unhealthy status, and recovers when it transitions back to an unhealthy one.
- Each agent should receive its own alert
- All four status rule templates use alert and recovery conditions
- The wide-window mitigation from https://github.com/elastic/integrations/issues/18885 is reverted