feat: normalize workspace agent session counts into a child table - #27952

Supersedes #27179 (same branch, squashed and rebased; the original PR had accumulated too much review noise).

Summary

Replaces the four fixed session_count_* columns on workspace_agent_stats with a normalized child table, workspace_agent_session_counts, keyed by app name, so any IDE or app can report session counts without schema or code changes.

Problem

Session counts were pinned to four hardcoded columns (vscode, jetbrains, reconnecting_pty, ssh). Every new IDE either mislabeled itself as one of the four or was dropped entirely, and supporting a new name meant a migration plus code changes at every layer.

Fix

• New workspace_agent_session_counts table (PK (workspace_agent_stats_id, app_name), created_at copied from the parent for windowed pruning, BRIN index), backfilled from the current stats buffer so rollups see no gap during upgrade.
• Agent API v2.11 adds a session_counts map to Stats, deprecating the fixed fields. Old agents' fixed fields are converted server-side.
• coderd/idemetadata holds the shared vocabulary: Normalize canonicalizes client-supplied names at ingestion, and Family groups names into bounded families for metric labels.
• Ingestion caps distinct app names at 64 per report; overflow aggregates under unknown and well-known names are never evicted. The agent applies a matching cap to concurrently active session types.
• POST /workspaces/{workspace}/usage and coder ssh --usage-app now accept arbitrary app names; the allowlist is replaced by ingestion-time normalization.
• Read-side queries (insights, deployment/agent stats) still pivot on the four well-known names. Family-based grouping at read time and richer client-reported names are follow-up work.

NOTE: The migration backfills the child table and rebuilds an index on workspace_agent_stats under lock. The stall is bounded by the stats retention window (~1 day of rows normally, up to 180 days on deployments where the purge never ran).

Mixed-version note: a new CLI passing a custom --usage-app value to an older coderd gets a 400, since old servers still validate against the fixed allowlist.

🤖 Generated with Claude Code