fix(native-profiler): prevent shell injection via device_id on iOS

[latekvo]

Summary

The iOS native profiler interpolated the caller-supplied device_id (udid) directly into shell-interpreted execSync template strings, so an unvalidated device_id could execute arbitrary shell commands.

In packages/tool-server/src/tools/profiler/native-profiler/platforms/ios.ts:

• execSync(\xcrun simctl spawn ${udid} launchctl list`)`
• execSync(\xcrun simctl listapps ${udid} | plutil -convert json -o - -`)`

device_id is declared z.string() with no format constraint and resolveDevice() returns it unchanged, so a value like `x"; touch /tmp/pwned #` runs as shell.

Fix

• Convert both call sites to execFileSync arg-arrays (no shell). The simctl listapps | plutil pipe is split into two in-process stages so nothing is shell-interpreted.
• Add a regression test asserting a hostile udid is only ever passed as a discrete argv element, with execSync mocked to fail loudly if it is ever reintroduced.

Context

This fix previously existed on an old branch but was lost when the code moved from tools/profiler/ios-profiler/ios-profiler-start.ts to native-profiler/platforms/ios.ts.

Verification

• npm run build (tsc) ✓ · typecheck:tests ✓ · prettier --check ✓ · eslint --max-warnings 0 ✓
• New test passes; reverting the fix back to execSync makes it fail (the payload booted"; touch /tmp/argent-pwned # reaches the shell command string), confirming it is a real guard.