kernel_copy tests: properly join background threads

[RalfJung]

This helps Miri because when there are threads just hanging in the background, we cannot check for memory leaks.

I think even without Miri this currently leads to 100% CPU load on one core for each of these threads: once the test function returns and sink gets dropped, the read in the thread will always immediately return 0, and then we just loop with that forever until the process exits.

[rustbot]

r? @Mark-Simulacrum

rustbot has assigned @Mark-Simulacrum.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: @ChrisDenton, libs
• @ChrisDenton, libs expanded to 8 candidates
• Random selection from Mark-Simulacrum, nia-e