Fix race condition with Nanny.on_exit_sync callback

[hendrikmakait]

Partially addresses #6625
We have a race condition with the Nanny.on_exit_sync callback, which might be executed after the _ongoing_background_tasks are already closed, resulting in an AsyncTaskGroupClosedError. In that case, something else is already closing the nanny, so we do not need to bother. Ideally, the on_exit and close logic would be generally cleaned up and avoid concurrent invocations, but that will likely need to be a more elaborate fix.

https://github.com/fjetter/distributed/runs/7090337573?check_suite_focus=true#step:11:3425

• Tests added / passed
• Passes pre-commit run --all-files

[github-actions]

Unit Test Results

See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests.

       15 files  ±0         15 suites  ±0   6h 43m 33s ⏱️ + 5m 40s
  2 908 tests ±0    2 823 ✔️ +3    81 💤  - 1  3 ❌  - 1  1 🔥  - 1 
21 533 runs   - 1  20 576 ✔️  - 1  952 💤 +2  4 ❌  - 1  1 🔥  - 1 

For more details on these failures and errors, see this check.

Results for commit 477e76c. ± Comparison against base commit 0f77c0d.

[quasiben]

The gpuCI failures are a known issue due problems with rmm/cuda-python. Feel free to merge in