[bug] Notification rule ANY to CRIT not working

__Steps to reproduce:__
1. Add Check (for every minute) and Notification endpoint

2. Add Notification rule `Critical` with condition: status changes from `ANY` to `CRIT`

3. Add Notification rule `Alive` with condition: status changes from `ANY` to `OK`

4. Stop `Telegraf` on monitored host
5. Wait for 2 minutes
6. Run `Telegraf` on monitored host
7. Wait for 2 minutes

__Expected behavior:__
Received 2 messages:
```
Host status is `crit`
Host status is `ok`
```

__Actual behavior:__
Received 1 message:
```
Host status is `ok`
```

__Environment info:__
* System info: `Linux 4.9.0-16-amd64 x86_64`
* InfluxDB version: `InfluxDB 2.0.3 (git: fe04d346df) build_date: 2020-12-15T01:00:16Z`

__Config:__
default

__Alert History:__
Checks:

Notifications:

__Conclusion:__
Something wrong happens with detecting `crit` -> `ok` transition

Start by reproducing the notification-rule scenario with a per-minute Check, a Notification endpoint, and Telegraf stopped and restarted as described. Trace how transitions from ANY to CRIT and from ANY to OK are evaluated, then use the Check and Notification alert histories to verify that both expected messages are delivered.

Written by the indexing model from the issue text.