refactor: move benchmarks to separate crate (#148)

## 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

```bash
# 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.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: beinan

Cargo.lock

@@ -3,5 +3,6 @@ members = [
     "crates/lance-graph",
     "crates/lance-graph-catalog",
     "crates/lance-graph-python",
+    "crates/lance-graph-benches",
 ]
 resolver = "2"

Cargo.toml

@@ -0,0 +1,20 @@
+[package]
+name = "lance-graph-benches"
+version = "0.1.0"
+edition = "2021"
+publish = false
+description = "Benchmarks for lance-graph (not published)"
+
+[dependencies]
+lance-graph = { path = "../lance-graph", version = "0.5.3" }
+lance = "1.0.0"
+arrow-array = "56.2"
+arrow-schema = "56.2"
+criterion = { version = "0.5", features = ["async", "async_tokio", "html_reports"] }
+futures = "0.3"
+tempfile = "3"
+tokio = { version = "1.37", features = ["macros", "rt-multi-thread"] }
+
+[[bench]]
+name = "graph_execution"
+harness = false

crates/lance-graph-benches/Cargo.toml

@@ -0,0 +1,28 @@
+# Lance Graph Benchmarks
+
+Performance benchmarks for the `lance-graph` crate.
+
+## Running Benchmarks
+
+From the repository root:
+
+```bash
+# Run all benchmarks
+cargo bench -p lance-graph-benches
+
+# Run specific benchmark
+cargo bench -p lance-graph-benches --bench graph_execution
+```
+
+## Benchmarks
+
+- **graph_execution**: End-to-end query execution benchmarks
+  - Basic node filtering and projection
+  - Single-hop relationship expansion
+  - Two-hop relationship expansion
+  - Tests with datasets of varying sizes (100, 10K, 1M rows)
+
+## Note
+
+This crate is not published to crates.io and is excluded from releases.
+It exists solely for performance testing during development.

crates/lance-graph-benches/README.md

@@ -47,13 +47,7 @@ unity-catalog = ["lance-graph-catalog/unity-catalog"]
 delta = ["dep:deltalake", "dep:url"]
 
 [dev-dependencies]
-criterion = { version = "0.5", features = ["async", "async_tokio", "html_reports"] }
 futures = "0.3"
 lance-arrow = "1.0.0"
-lance-index = "1.0.0"
 tempfile = "3"
 tokio = { version = "1.37", features = ["macros", "rt-multi-thread"] }
-
-[[bench]]
-name = "graph_execution"
-harness = false