feat: implement risk engine (guard, factors, and client integration)

[abiorh-claw]

Description

Implements the Risk Engine (Phase 1 & 2) for OmniClaw.
This system evaluates transactions based on risk factors (velocity, amount, new recipient) and assigns a Risk Score (0-100).

• Low Risk (0-20): Auto-Approved
• Medium Risk (20-80): Flagged → Creates 'Payment Intent' for HITL Review
• High Risk (80-100): Blocked

Key Features

• RiskGuard: Aggregates factors and enforces thresholds.
• RiskFactors:
  • AmountFactor: Exponential risk scaling for large amounts.
  • VelocityFactor: Checks transaction frequency spikes.
  • NewRecipientFactor: Higher risk for unknown addresses.
• Client Integration: pay() now catches RiskFlaggedError and automatically creates a pending PaymentIntent instead of failing.
• Trust-Awareness: Risk thresholds are dynamically adjusted based on ERC-8004 Trust Scores (if available).

Changes

• Added src/omniclaw/risk/guard.py and factors.py.
• Updated OmniClawClient.pay() to handle risk exceptions.
• Added examples/test_risk.py for verification.
• Added spec docs/specs/002_risk_engine.md.

Verification

• Ran examples/test_risk.py locally. Confirmed Allow, Flag (Intent Creation logic simulated), and Block scenarios.