Add optional comparison benchmarks that directly compare xtensor
performance against raw C++ implementations. This helps users
understand the performance characteristics and overhead of xtensor.

Changes:
- Add XTENSOR_BUILD_COMPARISON_BENCHMARKS CMake option (OFF by default)
- Add benchmark/comparison/ directory with BLAS1 comparisons
  - add_vector: z = x + y (vector + vector)
  - add_scalar: z = x + a (vector + scalar)
  - mul_scalar: y = a * x (scalar * vector)
- Each benchmark compares raw C++ vs xtensor (xarray, xtensor, noalias)

Usage:
  cmake .. -DXTENSOR_BUILD_COMPARISON_BENCHMARKS=ON
  make benchmark_comparison
  ./comparison/benchmark_comparison

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

Reduce code duplication and improve readability:
- Add helper functions for xtensor/xarray creation
- Add BENCHMARK_LOOP macro to reduce boilerplate
- Add REGISTER_BENCHMARK macro for standard sizes
- Rename: raw_cpp → std, shorten xtensor variants
- Add size 8 to benchmark ranges
- Use constexpr for scalar values

Changes:
- blas1_comparison.cpp: 211 → 197 lines (-14 lines)
- Add min_size=8, max_size=16384, multiplier=4 constants
- Simpler, more maintainable code

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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