feat(ethena): run Ethena + LlamaRisk backing checks independently

[spalen0]

Summary

Enables the previously-dormant LlamaRisk backing check as a second, fully independent USDe backing check, running alongside the Ethena transparency API check.

• Two independent checks — each provider is checked in isolation, so a failure or false positive in one source never suppresses the other.
• Provider-labelled alerts — every message is prefixed with the source that triggered it: [Ethena API] or [LlamaRisk]. No more ambiguity about which feed fired.
• Both checks compute (collateral + reserve fund) / supply, alerting CRITICAL if < 1 and HIGH if < 1.005.

Motivation

We received:

🔴 🚨 USDe NOT FULLY BACKED!
Backing Assets: $4,029,208,667.82
Total Supply: 4,031,893,924.09
Backing Ratio: 0.9993 (-0.07%)

This was a false positive: the active check used collateral-only, which hovers right around 1.0 and dips fractionally below it in normal operation. USDe holds its buffer in a separate reserve fund (~$62M) the check ignored. Including it → ratio 1.0154, comfortably backed. LlamaRisk independently reported 1.0424 at the same moment.

Changes

• Ethena check now includes the reserve fund (from Ethena's own /solvency/reserve-fund endpoint, which matches LlamaRisk's figure exactly — no cross-provider dependency).
• LlamaRisk check re-enabled as an independent second opinion with the same ratio/thresholds, plus its on-chain totalSupply() cross-validation.
• New main() runs both checks, each wrapped so one crashing can't stop the other.

Bug fixes in the LlamaRisk path

• Removed the dead self-comparison (supply vs itself → always 0).
• Fixed the on-chain totalSupply() double-divide by 1e18 in the log line.
• Guarded against get_tokens_supply() returning None.
• get_tokens_supply() now returns (None, None) on contract-init failure instead of bare None (which broke tuple unpacking).

Testing

• Live dry-run of both checks with Telegram stubbed: both run independently, log their provider, no false alerts (Ethena ratio 1.0154, LlamaRisk 1.0424).
• Forced sub-1.0 ratios to confirm each provider produces a correctly-labelled CRITICAL alert.
• ruff format / ruff check clean; tests/test_monitoring_config.py passes (9/9).

Docs

• Updated protocols/ethena/README.md and monitoring.yaml to document the two independent, labelled checks.

🤖 Generated with Claude Code