Phase 3 of the Scalable Approach for Adding New IDE Session Types RFC. The connection_type Postgres ENUM in connection_logs has the same fixed-set problem as the old session count columns: until it is gone, new IDE types are permanently recorded as vscode in the durable connection log.

Scope (RFC: "Migration plan (ENUM to TEXT)"):

• Migration: add type_text TEXT, backfill type::text, swap columns, drop the connection_type enum; remove the two ::connection_type casts in connectionlogs.sql. connection_logs is durable and can be large on busy deployments, so measure the backfill on a realistic dataset before deciding whether it needs batching
• Proto: add string type_str to Connection alongside the enum; new agents populate type_str with the normalized session type, the server prefers it and falls back to converting the enum (same pattern as Stats.session_counts in Normalize workspace agent session counts into a child table #27410)
• ReportConnection passes the normalized string through instead of classifying; db2sdk conversion becomes a passthrough
• Connection log UI filter groups by family via coderd/idemetadata; update connection-logs.md
• ~24 files, mostly mechanical (generated models, converters, enterprise connection log, SDK, frontend types)

Target: same release as #27410 (independent PRs). The client-side version gate (coder/vscode-coder#1044) points at the release containing this migration, so per-IDE names only start flowing once both the ephemeral and durable pipelines accept them; the gate is the safety net if this slips a release.

Done when

• connection_logs.type is TEXT and the connection_type enum is dropped
• Old agents (enum) and new agents (type_str) both log correctly
• A non-builtin session type appears in the connection log and its UI filter under the right family
• Release number recorded in Report the actual IDE (Cursor, Windsurf, ...) for session tracking vscode-coder#1044 for the client gate

Depends on #27410.

🤖 Created by Coder Agents on behalf of @EhabY.