feat: implement v2 3-tier architecture (mcpctl → mcplocal → mcpd)

- Rename local-proxy to mcplocal with HTTP server, LLM pipeline, mcpd discovery
- Add LLM pre-processing: token estimation, filter cache, metrics, Gemini CLI + DeepSeek providers
- Add mcpd auth (login/logout) and MCP proxy endpoints
- Update CLI: dual URLs (mcplocalUrl/mcpdUrl), auth commands, --direct flag
- Add tiered health monitoring, shell completions, e2e integration tests
- 57 test files, 597 tests passing

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/mcplocal/tests/metrics.test.ts

@@ -0,0 +1,93 @@
+import { describe, it, expect } from 'vitest';
+import { FilterMetrics } from '../src/llm/metrics.js';
+
+describe('FilterMetrics', () => {
+  it('starts with zeroed stats', () => {
+    const m = new FilterMetrics();
+    const stats = m.getStats();
+    expect(stats.totalTokensProcessed).toBe(0);
+    expect(stats.tokensSaved).toBe(0);
+    expect(stats.cacheHits).toBe(0);
+    expect(stats.cacheMisses).toBe(0);
+    expect(stats.filterCount).toBe(0);
+    expect(stats.averageFilterLatencyMs).toBe(0);
+  });
+
+  it('records filter operations and accumulates tokens', () => {
+    const m = new FilterMetrics();
+    m.recordFilter(500, 200, 50);
+    m.recordFilter(300, 100, 30);
+
+    const stats = m.getStats();
+    expect(stats.totalTokensProcessed).toBe(800);
+    expect(stats.tokensSaved).toBe(500); // (500-200) + (300-100)
+    expect(stats.filterCount).toBe(2);
+    expect(stats.averageFilterLatencyMs).toBe(40); // (50+30)/2
+  });
+
+  it('does not allow negative token savings', () => {
+    const m = new FilterMetrics();
+    // Filtered output is larger than original (edge case)
+    m.recordFilter(100, 200, 10);
+
+    const stats = m.getStats();
+    expect(stats.totalTokensProcessed).toBe(100);
+    expect(stats.tokensSaved).toBe(0); // clamped to 0
+  });
+
+  it('records cache hits and misses independently', () => {
+    const m = new FilterMetrics();
+    m.recordCacheHit();
+    m.recordCacheHit();
+    m.recordCacheMiss();
+
+    const stats = m.getStats();
+    expect(stats.cacheHits).toBe(2);
+    expect(stats.cacheMisses).toBe(1);
+  });
+
+  it('computes average latency correctly', () => {
+    const m = new FilterMetrics();
+    m.recordFilter(100, 50, 10);
+    m.recordFilter(100, 50, 20);
+    m.recordFilter(100, 50, 30);
+
+    expect(m.getStats().averageFilterLatencyMs).toBe(20);
+  });
+
+  it('returns 0 average latency when no filter operations', () => {
+    const m = new FilterMetrics();
+    // Only cache operations, no filter calls
+    m.recordCacheHit();
+    expect(m.getStats().averageFilterLatencyMs).toBe(0);
+  });
+
+  it('resets all metrics to zero', () => {
+    const m = new FilterMetrics();
+    m.recordFilter(500, 200, 50);
+    m.recordCacheHit();
+    m.recordCacheMiss();
+
+    m.reset();
+    const stats = m.getStats();
+    expect(stats.totalTokensProcessed).toBe(0);
+    expect(stats.tokensSaved).toBe(0);
+    expect(stats.cacheHits).toBe(0);
+    expect(stats.cacheMisses).toBe(0);
+    expect(stats.filterCount).toBe(0);
+    expect(stats.averageFilterLatencyMs).toBe(0);
+  });
+
+  it('returns independent snapshots', () => {
+    const m = new FilterMetrics();
+    m.recordFilter(100, 50, 10);
+    const snap1 = m.getStats();
+
+    m.recordFilter(200, 100, 20);
+    const snap2 = m.getStats();
+
+    // snap1 should not have been mutated
+    expect(snap1.totalTokensProcessed).toBe(100);
+    expect(snap2.totalTokensProcessed).toBe(300);
+  });
+});