Use bun instead of node which embeded in codegraph

[sopaco]

https://bun.sh/blog/bun-v1.4
Bun is the complete toolkit for building and testing full-stack JavaScript and TypeScript applications.
In the newest version it has a biggest jump in Node.js compatibility. It reduces idle CPU usage by 5x, reduces memory usage by up to 35%, and starts 50% faster on Linux.

[colbymchenry]

Thanks — Bun 1.4's numbers are genuinely impressive, and we did read that announcement with interest. Not planned, though, and it's worth recording why rather than just closing:

The bundled Node runtime is load-bearing, not incidental:

• node:sqlite is the sole database backend (WAL + FTS5, no native build step, no wasm fallback) — the entire "works on any machine with zero compilation" story hangs off that exact module in that exact runtime.
• The extraction kernel is a native N-API addon loaded per-platform by the bundled runtime; parse/store/query pools ride worker_threads; the daemon's process-lifecycle machinery (watchdogs, file watchers, socket handling) is validated per-platform against Node semantics — including some hard-won edge cases on Windows and Linux.

Swapping the embedded runtime re-opens every one of those invariants on three platforms. And the headline wins compound least for us: the MCP server is a long-lived daemon, so a faster cold start and lower idle CPU amortize to roughly nothing over a session — while indexing hot paths are dominated by the native kernel and SQLite, not JS runtime overhead.

If Bun's node:sqlite / N-API / worker_threads compatibility matures to true drop-in, the calculus could change and we'd revisit. Closing as not planned for now.