feat(safe): monitor yearn multisigs with proposer-based filtering

[spalen0]

Summary

• Adds 16 yearn-controlled multisigs (mainnet/base/katana) to the existing Safe queued-tx monitor, routed to a new YEARN_MS telegram destination. Source: engn33r/multisigs.csv. Other chains (Polygon, Arbitrum, Optimism, Fantom) intentionally excluded for now.
• Filters pending txs against a per-safe allow-list of recurring proposer addresses (YEARN_EXPECTED_PROPOSERS) so only unexpected proposers raise an alert. Allow-list discovered from the last ~50 executed txs per safe (≥20% share & ≥5 occurrences). On current pending state this filters ~70% (89/127).
• Splits safe/main.py 483 → 276 lines by moving static config (addresses, proposers, chain/api maps, proxy-upgrade selectors) into safe/addresses.py.
• Adds Katana (katana-main) to safe_apis, safe_address_network_prefix, and safe_network_to_chain_id alias map.

Telegram routing uses the standard TELEGRAM_CHAT_ID_YEARN_MS secret (a group chat, not a forum topic) — set the secret in repo settings once the group id is known.

Notes / follow-ups

• Some pending txs still alert because their proposer wasn't in the top-50 historic sample — e.g. SAM mainnet has a recurring 0x80c9aC86... not yet in the allow-list. Easy to extend after live alerts land.
• A few safes intentionally have no filter: V3 Deployer katana (only 2 historic txs, both pre-proposer-field) and Core Dev mainnet (48/50 historic txs pre-proposer-field). They alert on every queued tx as before.

Test plan

• Set TELEGRAM_CHAT_ID_YEARN_MS repo secret to the yearn_ms group chat id
• Verify the existing multisig-checker.yml workflow run (safe/main.py) succeeds end-to-end
• Confirm safe_apis["katana-main"] resolves a 200 from Safe (will be silent if 404)
• Watch live alerts for a few cycles; extend YEARN_EXPECTED_PROPOSERS with any obviously-recurring "unexpected" proposers
• Confirm yearn_ms group receives messages and non-yearn protocol alerts still route to their existing chats

🤖 Generated with Claude Code