Allow Smart Alerts to train on historical metric data collected prior to alert creation

Problem Statement:

Instana Smart Alerts based on adaptive thresholds require a fixed 14‑day training period starting from alert creation, regardless of how long the underlying entity has already been monitored.

This creates a significant limitation:

• Entities may have weeks or months of high-quality metric data
• Yet alerting teams are forced to wait an additional 14 days before adaptive alerts become reliable
• During this window, teams must either:
  • Operate without effective alerting, 
  • Fall back to static threshold alerts that are noisy, brittle, and often untrusted

This significantly slows down alerting configuration during:

• New service onboarding
• Production go-lives
• Incident-driven alert tuning
• Migrations to Instana Smart Alerts

Proposed Enhancement:

Allow Smart Alerts to optionally train on existing historical metric data for the monitored entity.

Possible approaches (you don’t need to prescribe implementation, just intent):

• Use the last N days (up to 14) of existing metric history as the initial training window
• Provide a toggle or option during alert creation:
  • “Use existing historical data for initial baseline”
• Clearly surface data-quality checks (e.g., insufficient history, topology changes)

Acceptance Criteria:

• If an entity has ≥14 days of continuous metric history:
  • Smart Alert becomes active with adaptive thresholds immediately or within hours
• If history is less than 14 days:
  • Fall back gracefully to existing behavior
• UI clearly indicates:
  • Whether historical data was used
  • Confidence level of the trained baseline