feat: agent effectiveness scoring — per-agent success rate and speed tracking

[rlaope]

Summary

Track per-agent success rate, retry count, and speed across sessions. This is the foundational feature that enables team auto-optimization (#5) and cross-project learning (#6).

Requirements

• src/observe/agent-scorer.ts — core scoring engine (agent ID → effectiveness metrics)
• src/observe/agent-storage.ts — persist per-agent stats to ~/.bestwork/agents/{agentId}.json
• src/harness/agent-metrics.ts — hook into agent completion events (PostToolUse)
• bestwork agents --scores CLI command to display rankings
• Enhance existing src/observe/effectiveness.ts with per-agent granularity
• Update hooks/hooks.json to capture agent completion events

Acceptance Criteria

• Each agent run records: agent name, task type, duration, success/fail, retry count
• bestwork agents --scores shows per-agent rankings (success rate, avg speed)
• Data persists in ~/.bestwork/agents/ across sessions
• Meeting logs include agent performance data
• Tests for scoring engine and storage

Complexity

3/5 — new metric types, storage pattern, hook integration

Dependencies

None (foundational)

Note

Issue #17 originally tracked this. This is the detailed spec.