feat: add record_task_outcome() API to RewardEngine #396
Description
Summary
RewardEngine provides multi-dimensional trust scoring but no API for recording task outcomes. Consumers must reimplement success/failure scoring math (severity-based penalties, diminishing returns, time-based recovery).
Proposal
Add task recording to RewardEngine:
- record_task(agent_id, outcome: success|failure, severity: float, metadata: dict)
- Built-in scoring: success = +boost with diminishing returns; failure = -10x severity
- Time-based trust recovery for idle agents (configurable rate)
- Threshold-based tier transitions (revoked/probationary/standard/trusted/verified)
- Persistent state via existing AuditChain
Why This Matters
Every toolkit consumer tracking agent trust needs to record task outcomes. The scoring math (diminishing returns, severity weighting, idle recovery) is universal — it shouldn't be reimplemented per consumer.
Real-World Evidence
ai-native-team daemon trust.py implements exactly this as a wrapper around RewardEngine. The wrapper adds ~150 lines of scoring logic that should be in the engine itself.