- Added pgo_training binary for PGO workload generation
- Covers k values from 3 to 25 as requested
- Tests multiple data distributions:
  - Uniform distributions  
  - Normal distributions with various parameters
  - Multimodal distributions (mixture of normals)
  - Skewed distributions (exponential)
  - Sorted data (timestamps)
- Tests various data sizes from 100 to 100,000 elements
- Added rand and rand_distr as dependencies for workload generation
- Handles edge cases with small datasets

- Created scripts/pgo-build.sh for automated PGO workflow
- Handles complete PGO pipeline:
  - Builds with instrumentation profile
  - Runs comprehensive training workloads
  - Merges profiling data using llvm-profdata
  - Builds final optimized binary with PGO data
- Supports multiple LLVM versions (looks for llvm-profdata variants)
- Platform-aware (detects OS and architecture)
- Includes error handling and coloured output
- Copies optimized binaries to target/pgo-optimized/
- Optional benchmark comparison via RUN_BENCHMARKS=1

- Created .github/workflows/pgo-release.yml for automated PGO releases
- Supports Linux, macOS, and Windows platforms
- Automatically builds PGO-optimized binaries on version tags
- Includes platform-specific LLVM tools installation
- Runs benchmarks comparing standard vs PGO builds
- Uploads PGO artifacts for each platform
- Creates GitHub releases with PGO-optimized binaries
- Can be manually triggered with workflow_dispatch
- Added rustup component add llvm-tools-preview to CI workflow
- Made system LLVM installations optional (continue-on-error)
- Updated Windows build to find and use Rust's bundled llvm-profdata
- Fixed benchmark execution for Windows platform
- CI now prioritizes Rust's bundled tools over system LLVM

- Added Profile-Guided Optimization section to README
- Documented expected performance improvements (10-30%)
- Provided build instructions using the automated script
- Explained what the PGO build process does
- Added usage instructions for different deployment scenarios
- Noted optimal use case alignment (k=3-25)

- Created benches/pgo_validation.rs for measuring PGO improvements
- Comprehensive benchmark suite covering k=3-25 range
- Tests multiple distributions:
  - Uniform distribution (common for binning)
  - Normal distribution (real-world data)
  - Multimodal distribution (cluster separation)
- Edge case testing:
  - Small k with large datasets
  - Large k with moderate datasets
- Uses criterion for statistical benchmarking
- Parameterized benchmarks for easy comparison