Investigate flaky Windows run_multithreaded_pe timeout

The Windows CI job on #1329 timed out while running `litebox_runner_windows_userland::run::run_multithreaded_pe`, despite the PR changing only `litebox_platform_linux_userland/src/lib.rs`.

In the [failed job](https://github.com/microsoft/litebox/actions/runs/34255366297/job/102159684113), 634 of 635 tests passed before `run_multithreaded_pe` hung after launching `litebox-broker-userland.exe`. Nextest terminated it at the 600-second timeout without additional stdout or stderr indicating the blocking operation.

The [failed-job rerun](https://github.com/microsoft/litebox/actions/runs/34255366297/job/102174294767) passed without code changes: `run_multithreaded_pe` completed in 6.474 seconds, and all 635 tests completed in 70.511 seconds. This suggests a timing-dependent hang or lost-progress condition in the multithreaded Windows runner path.

The test should complete reliably or fail with diagnostics that identify which broker, runner, or guest thread is blocked.

Start with litebox_platform_linux_userland/src/lib.rs and the litebox_runner_windows_userland::run::run_multithreaded_pe entry point. Compare the failed and successful Windows CI runs, focusing on the launch of litebox-broker-userland.exe and the absence of output after the hang. Done means the test completes reliably or reports which broker, runner, or guest thread is blocked.

Written by the indexing model from the issue text.