moonshine.rn: is_lfs_pointer_file depends on undocumented rg, and the iOS build lacks the LFS preflight guard #443

Two smaller findings from the Codex review of #441, both in packages/moonshine.rn.

1. is_lfs_pointer_file silently fails without rg

build-moonshine-android.sh:142:

if rg -q '^version https://git-lfs.github.com/spec/v1' "$file_path"; then

rg (ripgrep) is not declared as a dependency anywhere and is not present on a stock macOS or CI image. Under set -euo pipefail inside an if, a missing binary makes the condition false rather than aborting — so an unmaterialized LFS pointer is silently treated as a real file, and the guard added in #441 does not fire. The build then proceeds to the opaque CMake failure the guard exists to prevent.

Fix: use portable grep -q. Note the helper is duplicated, so both copies need it.

Regression check: with rg removed from PATH, pointer detection still returns 0.

2. The iOS build script has no equivalent guard

build-moonshine-ios.sh shares the same shape as the Android script — the same unconditional core/CMakeLists.txt source list and the same LFS flow — but has no post-pull check. It fails the same opaque way when the asset is unmaterialized.

Fix: add the equivalent guard before CMake runs.

Regression check: missing, pointer, override-supplied, and materialized cases each behave correctly.

Context

Neither is a regression from #441; both are pre-existing. The rg dependency is the more consequential of the two, since it can silently disable the guard that PR just added.

Related: #442 (setup.sh checks out a tag whose tree lacks the file), #437.