fix: call initSchema() in connectEngine() to auto-apply migrations on serve

[oyi77]

Problem

When gbrain serve starts, it calls connectEngine() → engine.connect() but never calls initSchema(). This means schema migrations (including forward-reference bootstrap columns like deleted_at, agent_name, source_id) are never applied on serve startup.

If a brain was created with an older gbrain version, serving with a newer version causes runtime errors:

column "deleted_at" does not exist

Root Cause

initSchema() was only called from gbrain init (3 call sites in init.ts), but not from the serve flow. The connectEngine() function in cli.ts creates the engine and connects, but skips schema initialization.

Fix

Add await engine.initSchema() to connectEngine() in cli.ts, right after connectWithRetry().

This is safe because:

• initSchema() is idempotent — it uses IF NOT EXISTS for all DDL operations
• applyForwardReferenceBootstrap() is a no-op on fresh installs and modern brains
• runMigrations() skips already-applied migrations
• Commands like gbrain init that call initSchema() themselves use an early return path before reaching connectEngine()
• The overhead is minimal (one probe query + conditional ADD COLUMN IF NOT EXISTS)

Testing

• Verified on an existing brain with v0.22 schema: 29 pending migrations applied successfully
• gbrain stats now returns correct data without errors
• gbrain doctor passes all checks

Fixes: #651

---

^{Need help on this PR? Tag @codesmith with what you need.}

• Let Codesmith autofix CI failures and bot reviews