fix(schema): backfill schema.py to match migrations.py reality (#691)

[AndriiPasternak31]

Summary

Brings src/backend/db/schema.py into alignment with src/backend/db/migrations.py, restoring it as a faithful reference for the database (Architectural Invariant #3 — "All table DDL lives in schema.py"). /validate-schema flagged 11 tables and 14 columns that existed in every running database (created by migrations) but were missing from schema.py.

The change is mechanical and additive — every CREATE TABLE / ALTER TABLE / CREATE INDEX statement in this PR was copied byte-for-byte from migrations.py. No migration changes. No code changes. Existing databases unaffected — every new statement uses IF NOT EXISTS.

Closes #691

What's added to schema.py

11 tables (each with the canonical DDL from its source migration):

• subscription_credentials (SUB-001) — declared before agent_ownership so the subscription_id FK target exists if FKs ever get turned on
• agent_notifications (NOTIF-001)
• subscription_rate_limit_events (SUB-003)
• slack_workspaces, slack_channel_agents, slack_active_threads (SLACK-002)
• telegram_bindings, telegram_chat_links, telegram_group_configs (TELEGRAM-001 / TGRAM-GROUP) — verified_email/verified_at and verified_by_email/verified_at rolled in from _migrate_access_control and _migrate_group_auth_mode
• whatsapp_bindings, whatsapp_chat_links (WHATSAPP-001)

14 columns rolled into existing CREATE TABLE bodies:

• agent_ownership: full_capabilities, max_backlog_depth, voice_system_prompt
• schedule_executions: source_user_id, source_user_email, source_agent_name, source_mcp_key_id, source_mcp_key_name, claude_session_id, queued_at, backlog_metadata, fan_out_id
• agent_schedules: webhook_token, webhook_enabled (feat: Webhook triggers for agent schedules (WEBHOOK-001) #291)

~22 indexes including the partials that were defined only in migrations:

• idx_executions_queued (partial WHERE status = 'queued')
• idx_executions_pending_retry (partial WHERE retry_scheduled_at IS NOT NULL AND status = 'pending_retry')
• idx_executions_fan_out
• idx_schedules_webhook_token (partial UNIQUE WHERE webhook_token IS NOT NULL)
• idx_agent_sharing_proactive (partial WHERE allow_proactive = 1)
• Plus table-local indexes for every new table

Why now

schema.py was a 70%-true reference — anyone reading it to learn the schema saw the wrong picture, and the architecture cross-reference to db/schema.py as canonical DDL was misleading. This PR closes that gap. It does not eliminate the two-source-of-truth pattern that lets drift recur — see "Follow-up" below.

Out of scope (intentional, per plan)

• No changes to migrations.py — migrations are append-only history.
• No code changes — routers/services already work against the live schema.
• No architecture.md update — Invariant Feature/vector log retention #3 is already documented.

Follow-up

A CI gate that fails PRs adding a migration without updating schema.py was approved at the autoplan final gate as a separate issue. Filing alongside this PR so the prevention path doesn't slip.

Test plan

• Plan's name-only diff script: missing tables: [], no missing cols, missing indexes: []
• Plan's strict DDL parity script: 0 missing, 0 different
• Migration ADD COLUMN cross-reference: every column added by a migration lives in schema.py
• tests/unit/test_migrations.py: 17/17 pass
• Smoke: backend boots against an existing ~/trinity-data/trinity.db (reviewer to verify; IF NOT EXISTS guarantees this is a no-op)

🤖 Generated with Claude Code

---

Migrated from #712 (head moved from fork to upstream as part of dropping the fork-based workflow). See #712 for original review thread.

[AndriiPasternak31]

Closing: this PR was opened to migrate #712 from fork to upstream, but #712 was already merged earlier today (commit d56012b). Branch AndriiPasternak31/claim-691 is already in dev via #712. No replacement needed.