fix(install): add prepare script so git installs build dist/

[johnkattenhorn]

Problem

Installing bmalph from git (e.g. npm install -g github:LarsCowe/bmalph or any fork thereof) leaves the CLI broken:

$ npm install -g github:LarsCowe/bmalph
added 24 packages in 8s

$ bmalph --version
/usr/bin/bmalph: line 1: syntax error near unexpected token

Root cause: npm's git-install path runs the prepare lifecycle script, not prepack. The current package.json only defines prepack (which runs when npm pack builds a tarball for registry publishing). So git installs skip the TypeScript build entirely and the installed package has no dist/ directory — bin/bmalph.js imports ../dist/... and crashes at launch.

Regular registry installs (npm install bmalph) work because the tarball published to npm already has dist/ baked in (via prepack).

Fix

Add a prepare script that builds when tsconfig.json is present (git-install case) and no-ops otherwise (tarball-install case):

"prepare": "[ -f tsconfig.json ] && npm run build || true",

The guard ensures the script is safe for both install paths:

• Git install: tsconfig.json is in the clone → build runs → dist/ populated → CLI works.
• Registry install: tsconfig.json is not shipped in the tarball (not in the files whitelist) → guard is false → no-op.

Why this matters for forks

Anyone forking bmalph for private or team use (pinning a commit for stability, or prototyping a change before upstreaming) hits this immediately. The fix is one line and is npm-lifecycle-standard.

Test plan

• npm install in a clone still works — prepare builds dist/ automatically.
• npm pack still produces a working tarball (prepack handles that path).
• Manual: npm install -g github:johnkattenhorn/bmalph#pr-prepare-script → bmalph --version prints version. Before this change: install succeeds but the binary is broken.
• Local test suite (npm test) — unchanged (4 pre-existing unrelated failures in installer.test.ts about .ralphrc template variants; not affected by this change).

Not in scope

• The 4 .ralphrc template-variant test failures I spotted during this work. Will follow up with a separate PR for that.