Implement Fuzzing Tests for VRM File Parser

# Implement Fuzzing Tests for VRM File Parser

**Labels:** enhancement, security, testing

## Problem
No fuzzing tests exist to discover potential crashes or security issues when parsing malformed VRM files.

## Current State
- No fuzzing tests for file parsing
- Unknown robustness against malformed files
- Potential security vulnerabilities undiscovered
- No systematic testing of edge cases

## Impact
- Potential crashes with malformed files
- Security vulnerabilities may exist
- Poor handling of edge cases
- Risk of exploits through crafted files

## Suggested Solution

### 1. Use Swift Fuzzing
Implement fuzzing using Swift's fuzzing capabilities or libFuzzer.

### 2. Generate Malformed VRM Files
- Invalid JSON structure
- Corrupted binary data
- Out-of-bounds indices
- Invalid buffer sizes
- Malformed GLTF structure
- Invalid VRM extensions

### 3. Test Parser Robustness
Ensure parser handles all malformed inputs gracefully without crashing.

### 4. Fix Discovered Issues
Address any crashes or vulnerabilities found through fuzzing.

### 5. Add Regression Tests
Add specific test cases for discovered issues.

## Example Fuzzing Setup
```swift
import Fuzzer

func fuzzVRMParser(data: Data) {
do {
let model = try VRMModel.load(from: data, device: device)
// Should either succeed or throw proper error
} catch {
// Expected for malformed data
// Should not crash
}
}
```

## Acceptance Criteria
- [ ] Fuzzing infrastructure set up
- [ ] Fuzzing tests run regularly
- [ ] All discovered crashes fixed
- [ ] Regression tests added for issues
- [ ] Parser handles all malformed input gracefully
- [ ] Security vulnerabilities addressed

**Priority:** Medium
**Effort:** Medium (2-3 weeks)

Start with the VRMModel.load(from:device:) entry point and the proposed Fuzzer-based Swift setup. Review how malformed JSON, binary data, GLTF structures, indices, buffer sizes, and VRM extensions are currently handled. Done means fuzzing runs regularly, crashes are fixed, and regression tests cover discovered issues.

Written by the indexing model from the issue text.