BUG: test flakiness observed in free-threaded Python

[neutrinoceros]

example logs
I'm just reporting this quickly to keep track on it. I'll dig down later and try to assess wether this is a concurrency issue or an actual memory leak, as the test is supposed to check for.

[neutrinoceros]

ok it's definitely flaky and not just on 3.13t: I just hit it on 3.14t too
https://github.com/pydata/bottleneck/actions/runs/27151228576/job/80142429565