feat(alerts): launch domain-isolated governance alerts worker v2

[0xPickles]

Closes dao-operations/triage#35

Summary

Replace the combined governance alerts bot with governance-alerts-bot-v2, a simpler Worker that independently monitors and delivers stYFI, veYFI, and yETH alerts.

The v2 Worker uses one Durable Object and one Telegram destination per domain:

• stYFI → t.me/styfi_alerts
• veYFI → t.me/veyfi_alerts
• yETH → t.me/yeth_alerts

The existing governance-alerts-bot remains a separate Worker and can continue serving the legacy combined channel during the transition.

Changes

• Split alert processing into three independently enabled Durable Objects with separate cursors, receipts, backoff state, and Telegram destinations.
• Replay each domain from its canonical start block using the same scanners, rendering, and delivery path used for live alerts.
• Add the approved message catalogue covering all supported stYFI, veYFI, and yETH actions.
• Add exact event-block account context to user alerts.
• Attribute supported direct and canonical Safe transactions without guessing through unsupported transaction wrappers.
• Support historical stYFI, veYFI, LLYFI, and yETH event shapes found during production replay.
• Evaluate yETH protocol changes at fixed daily checkpoints and emit alerts only for material changes.
• Set the yETH daily materiality threshold to 0.5 ETH.
• Remove the daily impact digest and the previous migration, warning, rollback-generation, lease, and general subrequest-governor machinery.
• Add a redacted authenticated /status endpoint and structured failure diagnostics.
• Retain simple resource bounds appropriate for the paid Cloudflare Workers plan:
  • 10,000-block log ranges
  • six ranges per domain run
  • RPC batches of 25 calls
  • five ordered Telegram messages per domain run
  • persisted Telegram retry_after handling
• Reserve disabled registry seams for teams, YBC, and DAO alerts.
• Add an exhaustive catalogue printer for reviewing canonical message output.
• Update the operator runbook, Worker documentation, message catalogue, replay procedure, and cutover instructions.

Architecture

governance-alerts-bot-v2 remains one Worker deployment, while domain state and failures are isolated across three Durable Object instances:

• alerts:styfi:v1
• alerts:veyfi:v1
• alerts:yeth:v1

A failure in one domain does not advance its cursor or prevent the other domains from running. Event receipts provide replay-safe delivery, subject to the documented at-least-once duplicate window if Telegram accepts a message before its receipt is persisted.

There is deliberately no automated health-monitor or operational-warning subsystem in this release. Failures remain visible through structured Cloudflare logs and the authenticated status endpoint.

Rollout and compatibility

• Introduces the permanent governance-alerts-bot-v2 Worker identity.
• Does not rename, overwrite, or adopt state from the existing governance-alerts-bot.
• Does not migrate the legacy singleton cursor or Telegram history.
• Preserves existing v2 Durable Object state and delivery receipts across redeployments.
• Keeps secrets in Cloudflare rather than source control.
• Commits all three production domain flags as enabled following successful replay testing.
• Allows the old combined channel to remain active during a deprecation period before the original Worker is disabled.

Verification

Completed successfully:

• TypeScript type checking
• ESLint
• Dependency-policy validation
• Full Vitest suite
• Focused scanner, transaction-attribution, accounting, RPC, and runtime tests
• Golden catalogue rendering
• Wrangler deployment dry run
• Exact production regression coverage for previously blocking stYFI, veYFI, and yETH transactions
• Public-RPC replay of veYFI from its previously blocked cursor through the confirmed head
• Full yETH replay from genesis through the confirmed head

The final replay fix passed 121 test files and 959 tests. No secrets are included in this change.