Create Automated Performance Benchmark Suite

# Create Automated Performance Benchmark Suite

**Labels:** low, testing, performance

## Problem
No automated performance tests exist to track regressions or validate performance claims in the README.

## Current State
- No automated performance tests
- Performance claims in README not backed by tests
- No baseline performance metrics
- Risk of performance regressions going unnoticed

## Impact
- Cannot track performance regressions
- Performance claims unverified
- Difficult to measure optimization impact
- No performance baselines for comparison

## Suggested Solution

### 1. Add XCTest Performance Tests
```swift
func testRenderingPerformance() throws {
let model = try loadTestModel()
let renderer = VRMRenderer(device: device)
renderer.loadModel(model)

measure(metrics: [XCTClockMetric(), XCTMemoryMetric()]) {
// Render 100 frames
for _ in 0..<100 {
renderer.draw(in: view, commandBuffer: commandBuffer,
renderPassDescriptor: descriptor)
}
}
}
```

### 2. Measure Key Metrics
- Frame time (average, p95, p99)
- GPU time
- Memory usage
- Draw call count
- Triangle count

### 3. Establish Baseline Metrics
Document expected performance for different model complexities.

### 4. Run in CI/CD
Add performance tests to CI pipeline to catch regressions.

### 5. Generate Performance Reports
Create automated reports comparing against baselines.

## Acceptance Criteria
- [ ] Performance tests for core operations
- [ ] Baseline metrics established
- [ ] Tests run in CI/CD pipeline
- [ ] Performance reports generated
- [ ] README claims validated by tests
- [ ] Regression detection automated

**Priority:** Low
**Effort:** Medium (1-2 weeks)

No file paths are identified. Start by locating the existing rendering entry points, the VRMRenderer usage, XCTest targets, README performance claims, and CI/CD configuration. Done means core-operation performance tests, baseline metrics, CI execution, regression reports, and validated README claims are all covered.

Written by the indexing model from the issue text.