fix(release): rebuild node-pty per target arch on macOS (closes #45)

[Ron537]

Fixes #45.

Summary

The v0.13.0 / v0.14.0 DPlex-x64.dmg is non-functional on every Intel Mac: opening a session fails with [Failed to start terminal], and the main-process log shows posix_spawnp failed from node-pty.

Root cause

electron-builder.yml had npmRebuild: false, which disables electron-builder's per-arch native-module rebuild step during packaging. With that flag off, the build process packages whatever node_modules/node-pty/build/Release/ happens to contain on the build host — and since our postinstall runs electron-builder install-app-deps once for the host arch, that directory only ever has binaries for the runner's arch (arm64 on macos-latest). Those arm64 binaries then get bundled into both the arm64 DMG (correct by coincidence) and the x64 DMG (broken — Intel can't exec arm64).

node-pty's native loader (lib/utils.js) probes build/Release/ → build/Debug/ → prebuilds/<plat>-<arch>/, so the stray host-arch build/Release/ wins over the per-arch prebuilds at runtime.

Fix

Remove npmRebuild: false. true is the default; with it enabled, electron-builder runs @electron/rebuild per target arch (clang on macOS cross-compiles x86_64 from Apple Silicon out of the box), so each arch's .app ships with the correct pty.node and spawn-helper. A comment in electron-builder.yml documents why this must stay enabled.

Alternatives considered and rejected

• Excluding build/** and falling back to the npm-shipped prebuilds/: tempting but the prebuilt spawn-helper ships with mode 0644. Confirmed locally that with only prebuilds present, posix_spawn fails with EACCES — same surface error as the arch-mismatch bug. Would have needed an afterPack chmod, i.e. binary fiddling.
• Universal .dmg: doubles download size and still needs per-arch rebuild internally.
• Per-runner split (macos-13 for Intel): GitHub is retiring Intel macOS runners; cross-compile from a single arm64 runner is the future-proof path.
• afterPack verification script: ~150 lines of CI surface to catch one regression class that's already documented inline in electron-builder.yml and visible in any PR diff. Not worth the maintenance.

Pre-release verification

Recommended before tagging v0.14.1:

gh workflow run release.yml -f dry-run=true
gh run download --name dplex-macos-latest
hdiutil attach DPlex-x64.dmg
file "/Volumes/DPlex"*/DPlex.app/Contents/Resources/app.asar.unpacked/node_modules/node-pty/build/Release/spawn-helper
# Expect: Mach-O 64-bit executable x86_64
hdiutil detach "/Volumes/DPlex"*

Test plan

• npm run lint — no new issues
• npm run typecheck — clean
• npm run test:unit — 325/325 pass
• Manual verification of the produced DPlex-x64.dmg on an Intel Mac (or via file inspection of the packaged binary) before tagging.

Diff size

+18 / −2 lines across 3 files. The actual fix is one line; the rest is the explanatory comment + version bump + changelog entry.

Thanks to @raphapizzi for the detailed report and root-cause analysis in #45 — it pointed straight at the right config flag.