Fix test_subscription flakiness

[lxfind]

Seems to fix it locally for me. Is this ok?

[gdanezis]

I totally approve of the had, and the comment about the mystery (on linux I cannot observe it).

[lxfind]

@gdanezis Looks like this fix is causing some tests run forever in CI. I wonder if you would be able to reproduce that in Linux and see why.

[gdanezis]

Looking into it.

[gdanezis]

Hey @lxfind - have a look at the modification. Comparing with t ie the instant that was returned, will always return true I think, so this was leading to an infinite loop (I think). Now I compare the amount of time elapsed since the start or last batch, which I think should increase.

However, I cannot know if this fixes the flakeyness on OSX :(.

[lxfind]

This still fails on my Mac.
Regardless what max delay I set, it always get hit many times, and if I print out the elapsed time, it's actually does print out the max delay each time, which is very strange. The only explanation is that tokio time/interval is buggy on Mac?
What I can do is to check on system time instead of tokio time, it seems to work.

[lxfind]

See #2094