fix: include last checkout verification before idle termination

[oliveigah]

closes #291

Currently, idle verification stops the pool if the first connection exceeds the idle timeout. I don’t believe this is the correct behavior, as it may lead to bugs like those described in the issue.

The general approach here was to add a last checkout timestamp to the pool state and verify it in the handle_ping callback.

While I’m not keen on adding this to the pool state just for idle verification, I couldn’t think of a better solution.

I’ll look into the possibility of adding a handle_ping callback specifically for the pool in nimble_pool, as the current handle_ping applies only to workers and doesn’t fully align with our use case. I think this would provide the semantics we need in Finch.

[oliveigah]

@mathieurousseau, I believe I’ve reproduced your issue in a unit test and have a solution in my fork.

If you could test it out in your code to see if it resolves the problem, that would be awesome!

[sneako]

Thanks a lot @oliveigah ! I'll wait to see if @mathieurousseau says this resolved the issue.