Fix #541: Track and display per-project agent time in analytics

[waleedkadous]

Summary

Adds Phase 1 agent time tracking to the analytics dashboard by summing consultation durations from MetricsDB per project and averaging by protocol.

Fixes #541

Root Cause

The analytics dashboard tracked wall clock time per protocol but had no visibility into how much actual agent compute time each project consumed. MetricsDB already records per-consultation duration_seconds with project_id and protocol fields, but no aggregation existed to surface this data.

Fix

• MetricsDB (metrics.ts): Added agentTimeByProtocol() method — a subquery groups consultations by (protocol, project_id), sums durations per project, then averages across projects per protocol
• Analytics server (analytics.ts): Extended ProtocolStats with avgAgentTimeHours, computed from MetricsDB data with graceful fallback if DB is unavailable
• Dashboard (api.ts, AnalyticsView.tsx): Display agent time alongside wall clock time in the Activity section's Projects by Protocol metrics

Test Plan

• Regression test: agent time appears in protocol stats when consultation data exists
• Regression test: null agent time when no consultation data for a protocol
• Regression test: graceful fallback when MetricsDB.agentTimeByProtocol() throws
• Build passes
• All 2055 tests pass (105 test files)
• Net diff: 112 LOC across 5 files

[waleedkadous]

Architect Review

Low-medium risk. Clean addition of per-project agent time tracking to analytics.

• SQL query is sound: inner GROUP BY (protocol, project_id) for per-project totals, outer AVG per protocol
• Graceful degradation via try/catch around MetricsDB — analytics never breaks
• Dashboard conditionally shows agent time only when data exists
• 3 new test cases cover the key paths (present, partial, DB failure)
• CMAP feedback addressed

Approved.

---

Architect review