[Metrics UI] Metric Threshold rule type should clear previous group cache on KQL query change

[Zacqary]

Metric Threshold rules store the previous list of detected groups in state.groups, in order to detect when groups unexpectedly disappear. To make sure we don't continue to alert on them when we do expect them to disappear, i.e. when the user edits the rule definition, we check to see if the groupBy (alert per) value has changed. But sometimes, changing the rule's KQL filter can also result in some groups disappearing.

See logic in https://github.com/elastic/kibana/blob/master/x-pack/plugins/infra/server/lib/alerting/metric_threshold/metric_threshold_executor.ts#L96-L98

const previousGroupBy = state.groupBy;
const prevGroups =
      alertOnGroupDisappear && isEqual(previousGroupBy, params.groupBy)
        ? // Filter out the * key from the previous groups, only include it if it's one of
          // the current groups. In case of a groupBy alert that starts out with no data and no
          // groups, we don't want to persist the existence of the * alert instance
          state.groups?.filter((g) => g !== UNGROUPED_FACTORY_KEY) ?? []
        : [];

Note that we are only checking if the groupBy value has changed — if it has changed, prevGroups evaluates as an empty array instead of the stored state.groups. This resets the cache, as prevGroups gets merged with the currently detected groups to get saved to state.groups.

Acceptance criteria

• Store the rule definition's filterQuery in rule state
• In addition to checking if groupBy has changed before building prevGroups, also check to see if the filterQuery has changed

[elasticmachine]

Pinging @elastic/logs-metrics-ui (Team:logs-metrics-ui)