Random failures to remove /tmp/vsc-coverage-

We spotted an increasing number of CI failures using vscode-test that seem to be flaky and likely caused by some timing issues. Example of run at https://github.com/ansible/vscode-ansible/actions/runs/21067364468/job/60587877011#step:16:779 exact error listed below, error that is causing the entire pipeline to fail.

Error: ENOTEMPTY: directory not empty, rmdir '/tmp/vsc-coverage-8b22aabc-fead-4441-9101-bbff43f023b2'

Some change is needed in order to prevent this from happening. Sadly we also discovered some time ago that there is no way to configure coverage options inside .vscode-test.mjs files as they only allow vscode-test specific settings and mocha settings but not settings for the coverage framework (only cli args work and they are very limiting).

The code comes from https://github.com/microsoft/vscode-test-cli/blob/044359f9cebf187b5702a768aff5d4640a152632/src/cli/coverage.mts#L23
and apparently it should be possible to configure tempDirectory for coverage but is not.

Start with src/cli/coverage.mts, especially the coverage temporary-directory handling linked in the issue, and review how .vscode-test.mjs and CLI arguments configure coverage. Reproduce or inspect the reported CI failure and determine how tempDirectory can be configured or cleanup made reliable. Done means vscode-test coverage no longer intermittently fails with ENOTEMPTY.