[telemetry] Analyze AI/agent impact on migration to custom templates

## Problem

Known-template usage is declining while agent usage and custom template is growing. We suspect the two are related — that users working through an agent are more likely to build a custom template or use their own codebase — but we have no data to confirm it.

This matters because it changes where we invest: if agents are driving users away from known templates, then template investment and agent investment should be weighted differently.

## What we'll do

Cross the initialization cohort with the execution environment over time, and look for whether agent-driven sessions skew toward custom templates and own-codebase cohorts.

Depends on:
- Cohort analysis by project initialization type
- Split user tables for local / CI/CD / agent

## Done when

We can state whether agent users are more likely to be on custom templates or their own codebase, and by how much.