[Vincent-520]

Five commits: the feature the branch is named for, and the four fixes this branch
needed before it would build at all. The build fixes are independent of the i18n
and can be taken on their own if you would rather split them.

1. Localise the app, and add Simplified Chinese (992b44b)

Every user-facing string now goes through a String Catalog.

• Localizable.xcstrings — 511 keys, en as the source language so an
  untranslated key falls back to itself, all 511 translated to zh-Hans.
• InfoPlist.xcstrings — the local-network prompt. Its English is spelled out
  in the catalogue as well, because its key is an Info.plist key rather than a
  sentence: without that entry the compiled en.lproj maps the key to itself
  and the prompt reads NSLocalNetworkUsageDescription.
• CFBundleLocalizations in Info.plist, and a Settings › Language screen that
  writes AppleLanguages so a language can be pinned. The screen says when the
  choice needs a relaunch to be seen, and says nothing when it does not.
• Localization.swift adds localizedKey(_:) and localizedString(_:); the
  shared components in Theme.swift now take LocalizedStringKey.

Most of the diff is the 265 places SwiftUI does not localise on its own —
anything the compiler sees as a String: a ternary, a concatenation, a value
out of a computed property, a parameter typed String. A sentence with a value
in the middle is written as one literal with an interpolation (Version %@),
never as "Version " + x, which would put the untranslated half inside the key.

Deliberately left alone: log lines, shell commands, paths, and the names of
things Android reports (app labels, ABIs, engine names). Those are evidence, not
sentences.

2. Four fixes that were stopping the IPA build

bb8d4ea added embedded MoltenVK and build-ipa.yml was not updated with it, so
every cold-cache run has failed since. Upstream's last green run predates that
commit, and its cache carries a build/ios-arm64/lib from before the dependency
existed — which is why this only shows on a fresh cache.

• 5071977 Build MoltenVK, in the layer the cache covers, with its script in
  that cache's key. The key goes to qemu-v2 because a v1 lib/ holds no
  MoltenVK and must not be restored.
• 8c0692d Install qemu-img — project.yml's pre-build step creates the
  firmware and userdata seeds with it and a fresh runner has none — and build
  libhusk_rppairing, which the app links (-lhusk_rppairing) and which
  nothing staged.
• 194c212 check_dualmap called sys_icache_invalidate directly instead of
  the portable flush_icache() the file already defines a few hundred lines
  above. The block is under #if defined(__aarch64__), which is true for the
  aarch64-linux-gnu build in tests/translation-layer/run.sh, so the test job
  died on an implicit declaration.
• acc2a23 With that compiling, the aarch64 test link stopped on
  tl_find_stikdebug_prewarmed, which check_dualmap calls and which lives in
  husk-tl-load.c. That file is the Apple-side loader — it includes
  mach/mach_time.h unguarded — so the host tests deliberately do not link it.
  Stubbed to return NULL, which is what the real lookup answers on a machine with
  no StikDebug, so the dualmap check still reports skip for the honest reason;
  the harness asserts exactly that. The generated stub file is renamed
  noscan.c → nostubs.c now that it holds two unrelated stubs.

Verification

CI is green on this branch, both jobs — run
37600731806,
including a from-scratch QEMU + MoltenVK + libhusk_rppairing layer.

The IPA that run produced carries zh-Hans.lproj/Localizable.strings with all
511 entries, zh-Hans.lproj/InfoPlist.strings, and
CFBundleLocalizations = [en, zh-Hans], with every other bundle check
scripts/package_ipa.sh makes passing.

[Leviidev]

Heavy conflicts, I prefer to add this feature myself in 1.0.0