timer: fix cancel in periodic timer callback

[brain5lug]

Fix new timers unable to set system timer callback
if there was removed timer with low new expire
time which was removed in their seastar callback

Please comment bad index calculated of timer then timer
is removed.

[avikivity]

Please add a test that demonstrates the bug (fails before the fix, passes after the fix).

[nyh]

I don't understand this comment. Why do you need the sleep_for()? the waiting for p4's future will wait for more than 100ms anyway.
Or maybe what you're saying is that somehow the cancel() broke this sleep_for() and before this patch, the sleep_for would never finish? If this is the case, please a comment saying this, and not some statement about "long 3rd party unstoppable calculation", whatever that means.

[brain5lug]

Added explicit comment that this line is for reactor delay

[nyh]

Please add a test that demonstrates the bug (fails before the fix, passes after the fix).

I see he did add a test, but never explicitly said that this test reproduces the bug - fails before the fix, and passes afterwards.
I would suggest to please:

1. Open a github issue about the bug. Explaining exactly what the bug is and what its symptoms were (but not how to fix it).
2. Above the test, add a comment explaing that it reproduces the issue (mentioning the issue number).
3. Add to the pull request a "Fixes #..." line - which will cause the issue to be automatically closed when the issue is merged.

Thanks.

[brain5lug]

Please add a test that demonstrates the bug (fails before the fix, passes after the fix).

The test has been already added. Probably it requires some improvements. Currently the test hangs forever (as reactor hangs) without the fix and passes the test if fix is applied. The fix is for one bug. I suspect another bug. I can prepare branch with assert which shall prove it exists.