refactor: move benchmarks to separate crate

Move benchmarks from lance-graph to a new lance-graph-bench crate
to exclude them from release binaries and reduce build times.

Changes:
- Create new crate: lance-graph-bench (publish = false)
- Move graph_execution benchmark to new crate
- Remove benchmark-only dependencies from lance-graph:
  - criterion (benchmarking framework)
  - lance-index (unused)
- Update workspace to include lance-graph-bench
- Add README for benchmark crate

Benefits:
- Smaller release binaries (benchmark code excluded)
- Faster production builds (benchmarks only built when needed)
- Cleaner separation of concerns
- All existing tests still pass (295 tests)

Usage:
  cargo bench -p lance-graph-bench

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-bench",
 ]
 resolver = "2"

Cargo.toml

@@ -0,0 +1,20 @@
+[package]
+name = "lance-graph-bench"
+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-bench/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-bench
+
+# Run specific benchmark
+cargo bench -p lance-graph-bench --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-bench/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