CodeGraph CLI MCP Server

# codegraph-graph — I/O Batching Enhancements This crate now includes an adaptive I/O batching layer to reduce syscall overhead and improve end-to-end latency for graph storage backed by RocksDB. ## Highlights - Read coalescing for `get_node` via a background aggregator that batches requests within a short coalescing window. - Adaptive write batching with dynamic thresholds tuned by EWMA of flush latencies and a time-based flush fallback. - Sequential access optimizations for prefix scans with `iterator_cf_opt` and 2MB readahead. - Conservative defaults that preserve safety (WAL enabled; `sync=false` for throughput). ## Key Components - `io_batcher::ReadCoalescer`: Groups concurrent node reads and serves them in batches. - `io_batcher::WriteBatchOptimizer`: Tunes write-batch operation thresholds to meet latency targets. - `storage::HighPerformanceRocksDbStorage`: Integrates batching and sequential-scan optimizations. ## Configuration `BatchingConfig` (internal default) controls: - `max_read_batch` (default 256) - `read_coalesce_delay` (default 300µs) - `initial_write_ops_threshold` (default 1000) - `min_write_ops_threshold` / `max_write_ops_threshold` - `write_flush_interval` (default 5ms) - `target_flush_latency_ms` (default 2.0ms) These defaults are tuned for low latency under mixed workloads and can be adjusted if a configuration surface is later exposed.