fix(build): make native ABI switching interruption-safe

[AmagiDDmxh]

Why

Spool keeps one installed better-sqlite3 binary and switches it between the Node and Electron ABIs. Two parts of that workflow were costly or fragile:

• The root pretest rebuilt the Node addon unconditionally, even when the installed binary already matched the current Node ABI.
• with-electron-native.mjs used synchronous child processes, so an interrupt could terminate the wrapper before its Node-ABI restoration completed.

What changed

• Probe better-sqlite3 with a real in-memory query before rebuilding for Node; rebuild only when the load/query fails.
• Run the Electron rebuild, wrapped command, and Node restoration asynchronously.
• Forward SIGHUP, SIGINT, and SIGTERM to the active child while keeping the wrapper alive long enough to restore the Node ABI.
• Preserve signal-style exit codes after restoration.
• Remove obsolete npm runtime/target environment overrides from the Node rebuild.
• Update contributor and development documentation to match the managed ABI workflow.

Verification

• A matching Node ABI took 1.91 seconds through pnpm run ensure:native:node and skipped compilation.
• A real Electron-ABI switch followed by Ctrl-C completed the Node rebuild; better-sqlite3 then loaded successfully under Node ABI 137 and executed an in-memory query.
• Both scripts passed node --check and focused Oxlint.
• Workspace typecheck, type-aware lint, core tests (417 passed, 1 skipped), and app tests (450 passed) completed successfully.
• The full local pnpm check reached the CLI suite, where 56/60 tests passed; four multi-process pin/unpin tests exceeded their existing 5-second timeout on this host. No timeout was changed or relaxed; PR CI remains the confirmation for those tests.