feat(telegram): route operational errors to a dedicated errors channel

[spalen0]

Problem

Operational errors are spamming the main per-protocol alert groups. Example reported on the aave topic:

GraphQL error in response: [{'message': 'bad indexers: {0x090f…: BadResponse(400), …}'}]

This came from an inline error send in protocols/aave/proposals.py that's caught and reported without crashing, so the run_with_alert crash wrapper never handled it — it went straight to the aave group.

What this does

Adds a dedicated errors channel so transient failures (GraphQL/fetch errors, retries, script crashes) route to one chat instead of the main groups.

Mechanism (utils/telegram.py)

• New send_error_message(message, protocol) + ERROR_CHANNEL = "errors" key, reusing the existing channel-routing scheme — configure entirely via env vars.
• Each message is prefixed with a [protocol] label so the merged feed shows which monitor produced it.
• Safe fallback: if no errors destination is configured, it routes to the protocol's own channel exactly as before. Backward-compatible — nothing changes until the env var is set.
• Always plain-text + silent by default.

Configuration (see .env.example)

# Pick ONE (topic takes precedence)
TELEGRAM_TOPIC_ID_ERRORS=321          # forum-style topics group
TELEGRAM_CHAT_ID_ERRORS=-1001234567   # or a standalone chat
TELEGRAM_BOT_TOKEN_ERRORS=...         # optional, falls back to DEFAULT bot

Call sites rerouted (16 files)

• All unhandled crashes — utils/runner.py (covers every script via run_with_alert).
• Reported case — aave/proposals.py (GraphQL error + Graph-API-retry failure).
• Inline operational errors — fluid, maker, lido, strata, yearn (Envio), timelock, morpho (fetch/GraphQL failures), and the send_alert(LOW, …failed) diagnostics in ethena, usdai, apyusd, infinifi, maple, rtoken.

Left in main groups (real protocol signals, not noise)

morpho 🚨 No vaults data found, ustb CRITICAL zero-price, ethena MEDIUM stale-data alerts.

Bonus cleanup

The timelock Envio-unreachable handler was fanning the error out to every monitored protocol's group (a big spam source) and referenced an undefined protocol on the GraphQL-error path (latent NameError). Both now collapse to a single labelled errors-channel send.

Testing

• ruff check ✅ / ruff format (no changes) ✅
• py_compile on all changed files ✅
• mypy — no errors in the changed files (pre-existing repo errors are unrelated)

🤖 Generated with Claude Code