[ML] Initial Alerting and Action integration

[darnautov]

Alerting and Actions is a centralized tool to manage complex conditions and invoking actions when those conditions are met. Introducing an ML alerts integration should provide a straightforward way to set up conditions for existing Anomaly Detection jobs and kick-off alerts in the job wizards. This will replace the current integration with Watcher.

• Implement a new Alert type for Anomaly detection ([ML] Anomaly Detection alert type #89286)
• Pre-populate the Alert type template in the Anomaly Detection job wizard ([ML] Anomaly Detection alert initialisation from the ML app  #91283)
• Initialize Anomaly Detection alert from the job list ([ML] Anomaly Detection alert initialisation from the ML app  #91283)
• Remove ability to create Watches for Anomaly Detection jobs ([ML] Anomaly Detection alert initialisation from the ML app  #91283)
• Avoid notification about the same alert ([ML] Prevent duplicate notifications about the same anomaly result  #91485)
• Adjust time range based on the check interval and the bucket span ([ML] Prevent duplicate notifications about the same anomaly result  #91485)
• Add an alert param for filtering interim results ([ML] Anomaly Detection alert initialisation from the ML app  #91283)
• Show an info callout about the result bucket span used based on the job selection ([ML] Fix Anomaly detection alert condition validation  #92377)
• Warn if datafeed is not started for selected jobs [ML] Fix Anomaly detection alert condition validation  #92377
• Replace context.kibanaBaseUrl variable with kibanaBaseUrl provided as a build-in variable

[elasticmachine]

Pinging @elastic/ml-ui (:ml)

[richcollier]

Had a discussion with @darnautov today on the following topics:

• interim results (how to handle)
• multi-bucket anomalies (option to include/exclude)
• alert config being "job aware" (notify user that you cannot do an influencer alert on a single metric job, for example)
• do we warn users if ML Custom Rules or Calendars may impede alert firing?
• do we allow user to control alert output text format (i.e. include actual/typical, etc.)
• should we consider an alert "tester" to reveal how many alerts would have been fired over the last X days (similar to SIEM)
• should we consider a "composite" alert type ((JobA>90 AND JobB>75) OR JobC>25) and/or weighted combos (0.8JobA + 0.5JobB +0.1*JobC)>X

[darnautov]

7.13 Meta issue #96381