feat(mcp): Add memory monitoring for long-running processes

[prosdev]

Problem

No memory monitoring for long-running MCP processes. While circular buffers exist, there's no usage tracking or automatic restart thresholds. AI tools become unresponsive after heavy use due to unbounded memory growth.

Current state:

• Circular buffers exist but no memory monitoring
• No automatic restart thresholds
• No heap usage tracking
• Production MCP servers can grow unbounded

Impact

• Reliability: MCP server becomes unresponsive after heavy use
• User Experience: AI tools slow down or crash during long sessions
• Production: No visibility into memory issues
• Success Criteria: <500MB steady state memory usage

Solution

Add memory monitoring with automatic restart thresholds:

1. Track heap usage periodically (every 30s)
2. Log warnings when approaching thresholds
3. Emit health check failures when over threshold
4. Optional: Graceful restart when critical threshold hit

Implementation approach:

// Memory monitoring service
interface MemoryStats {
  heapUsed: number;
  heapTotal: number;
  rss: number;
  external: number;
}

class MemoryMonitor {
  private checkInterval = 30000; // 30s
  private warnThresholdMB = 400;
  private criticalThresholdMB = 500;
  
  start() {
    setInterval(() => {
      const mem = process.memoryUsage();
      const heapMB = mem.heapUsed / 1024 / 1024;
      
      if (heapMB > this.criticalThresholdMB) {
        logger.error(`Memory critical: ${heapMB}MB`);
        // Emit health check failure
      } else if (heapMB > this.warnThresholdMB) {
        logger.warn(`Memory warning: ${heapMB}MB`);
      }
    }, this.checkInterval);
  }
}

Acceptance Criteria

• Memory usage logged every 30s in debug mode
• Warning emitted at 400MB heap usage
• Critical error at 500MB heap usage
• Health check reports memory status
• Circular buffer limits enforced (already exists)
• No memory leaks in 24h stress test

Files to Create/Modify

• packages/mcp-server/src/monitoring/memory-monitor.ts - New monitoring service
• packages/mcp-server/bin/dev-agent-mcp.ts - Wire up monitor
• packages/mcp-server/src/adapters/built-in/health-adapter.ts - Report memory
• Tests for memory monitoring

Priority: P1 - Production stability
Part of: #104 - Performance & Reliability Critical Path

Future Enhancement

Consider automatic graceful restart when hitting critical threshold (v2):

• Notify client of restart
• Save current state
• Spawn new process
• Transfer connections