[Epic] Redesign azd telemetry: retention, user surfaces, cohorts, and dashboards

## Goal

Improve azd telemetry so we can make design decisions from evidence, and grow adoption.

Our cooked tables and Power BI dashboards were built around templates, provision, and deploy. That no longer matches how azd is used:

- Most traffic now comes from **CI/CD and agents**, but MAU counts local/desktop only, so our headline number describes a shrinking slice of our users.
- Users increasingly start from a **custom template or their own codebase**, so the template pivot our dashboards depend on is going stale.
- **New features ship with no way to see whether anyone uses them**, because reporting is provision/deploy shaped.
- **Raw events expire after 90 days**, so no long-horizon question can be answered.
- **When telemetry breaks we find out late**, by noticing a chart looks wrong.

This epic tracks the work to close those gaps.

## Workstreams

**Compliance** — tracked in the dedicated GDPR telemetry epic
- #9801 GDPR classification and privacy enforcement for azd telemetry

**Data foundation**
- #9387 Retain raw events into cooked tables and share Kusto queries as team notes
- #9388 Split user tables: local (MAU), CI/CD, and agent, plus a combined total
- #9389 Move to year-over-year metrics
- #9390 Cohort analysis by project initialization type

**Dashboards**
- #9392 New azd data dashboard, separate from the template dashboard
- #9393 Power BI dashboard for feature usage outside provision and deploy
- #9394 Categorize builder extension metrics
- #9395 Agent/MCP telemetry: cooked tables, data quality, and dashboard cleanup

**Analysis and reliability**
- #9391 Analyze AI/agent impact on migration to custom templates
- #9396 Telemetry validation and monitoring

## Related

- #9345 — review and redesign azd telemetry end-to-end, and close the gaps

Kusto functions and Power BI models live in the internal `azure-dev-tools` repo under `product-telemetry/azd/`.