feat: add visual resources and Python examples

Visual Resources:
- Create comprehensive architecture diagram (architecture.svg)
  * Shows all modules: Elementwise, Reduction, GEMM, Attention, etc.
  * Illustrates technology stack: C++20, CUDA 12.4+, Python bindings
  * Uses color-coded gradients for different component types

- Create GEMM performance chart (performance-chart.svg)
  * Visualizes 7-step optimization journey
  * Shows performance progression from 0.5 to 70+ TFLOPS
  * Includes speedup annotations and status indicators
  * Highlights key insight: 100× speedup with Tensor Cores

- Design high-quality logo (logo-large.svg)
  * Modern GPU chip design with Tensor Core grid
  * Performance arrow showing 140× speedup
  * Professional gradient colors (CUDA green to blue)
  * Clean typography with tagline

Python Examples:
- Create GEMM benchmark Jupyter notebook
  * Performance comparison with PyTorch/cuBLAS
  * Multiple matrix size benchmarks
  * Visualization with matplotlib
  * Detailed correctness verification
  * Summary statistics and analysis

- Add advanced usage examples script
  * Elementwise operations (ReLU, Sigmoid)
  * Reduction operations (Softmax)
  * GEMM variations (square, non-square)
  * Batch processing patterns
  * PyTorch autograd integration
  * Performance profiling utilities

These additions significantly improve:
- Documentation visual appeal
- User onboarding experience
- Performance understanding
- Practical usage examples

Author: shijiashuai