Fix ARM64 CI flaky test output ordering

[zyla]

Summary

• Move stream handler waits (stdout/stderr) to before status logging in src/App.hs, ensuring all subprocess output is fully flushed before "success" or "skipped" messages are written
• Fixes a race condition where the main thread could log status messages before output handler asyncs finished reading subprocess output, causing non-deterministic line ordering visible in golden tests on ARM64

Root Cause

After waitForProcess returns, the main thread immediately writes status messages (e.g. logInfo "success") via toplevelStderr. However, the stdoutHandler async (reading subprocess stdout via toplevelStdout) was only awaited later. Both handles write to the same merged pipe in test mode, so on ARM64 (with different scheduler behavior) the status message could appear before the last subprocess output line.

Test plan

• All 34 tests pass with SKIP_SLOW_TESTS=1 SKIP_S3_TESTS=1 stack test
• Verify ARM64 CI no longer produces flaky golden test failures

🤖 Generated with Claude Code