refactor: move benchmarks to separate crate

[beinan]

Summary

Moves benchmarks from lance-graph to a new lance-graph-benches crate to exclude them from release binaries and reduce build times for production code.

Changes

New Crate: lance-graph-benches

• Created dedicated benchmark crate with publish = false
• Moved graph_execution benchmark from lance-graph
• Added README with usage instructions

Cleaned Up lance-graph

• ✅ Removed benchmark-only dependencies:
  • criterion (benchmarking framework)
  • lance-index (unused)
• ✅ Kept test dependencies:
  • tempfile (used by tests)
  • lance-arrow (used by tests)
  • tokio, futures (used by tests)
• ✅ Removed [[bench]] section from Cargo.toml

Workspace

• Added lance-graph-benches to workspace members

Benefits

✅ Smaller Release Binaries - Benchmark code excluded from published crates
✅ Faster Production Builds - Benchmarks only compiled when explicitly requested
✅ Cleaner Separation - Dev/benchmark code separate from production code
✅ All Tests Pass - 295 unit tests still passing
✅ Benchmarks Work - Verified compilation and structure

Usage

# Run all benchmarks
cargo bench -p lance-graph-benches

# Run specific benchmark
cargo bench -p lance-graph-benches --bench graph_execution

# Build production crate (benchmarks NOT included)
cargo build -p lance-graph

Testing

• ✅ cargo check -p lance-graph - success
• ✅ cargo build -p lance-graph - success
• ✅ cargo test -p lance-graph --lib - 295 tests pass
• ✅ cargo check -p lance-graph-benches - success
• ✅ No benchmark code in production binary

File Structure

crates/lance-graph-benches/
├── benches/
│   └── graph_execution.rs
├── Cargo.toml (publish = false)
└── README.md

This follows the common Rust pattern of separating benchmarks into dedicated crates when they have heavy dependencies or shouldn't be included in releases.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!