Add benchmark for query system overhead in eure-json

[ryo33]

Add criterion benchmarks to measure the overhead of the query-flow system
by comparing:

1. Query pipeline: Input -> CST -> EureDocument -> JSON via query runtime
2. Manual pipeline: Direct function calls without query system

Key findings from initial benchmarks:

• New runtime overhead: ~6.6x for small, ~2x for medium, ~1.16x for large inputs
• Query cache hit: ~43x faster than manual (660ns vs 28µs for small input)

Benchmarks include:

• full_pipeline: End-to-end conversion comparison
• query_reuse: Query execution with pre-built runtime (shows cache benefits)
• parse_phase: CST parsing performance
• document_phase: EureDocument construction performance
• json_phase: JSON conversion performance

Run with: cargo bench -p eure-json --bench conversion