Fix keepalive readiness deadlock when sub-millisecond time remains

[Copilot]

The macOS CI job timed out because ntex-util services::keepalive::tests::test_ka hung indefinitely (SLOW [>420.000s] until the 10-minute step timeout).

KeepAliveService::ready() re-arms its Sleep with the time remaining until expiration, truncated to whole milliseconds. now() is a low-resolution (5ms) clock, so the remainder is frequently sub-millisecond, producing Sleep::reset(Millis(0)). TimerHandle::update_timer treats 0 as "clear the timer", leaving the sleep permanently elapsed — the subsequent poll_elapsed returns Ready and registers no waker, so the dispatcher parks forever and readiness never resolves.

Changes

• ntex-util/src/services/keepalive.rs
  • Re-arm the sleep with a minimum of Millis(1) so a waker is always registered.
  • Replace the single re-poll with a loop, so ready() only returns Ok once the timer is actually pending (or Err if the keep-alive expired in the meantime).
  • Add test_ka_sub_millis, which forces the sub-millisecond state and asserts the timer is re-armed after ready().

let expire = expire - now;

// sleep must be reset to non zero duration,
// otherwise it stays in elapsed state and waker
// never gets registered
let expire: u32 = expire.as_millis().try_into().unwrap_or(u32::MAX);
self.sleep.reset(Millis(expire.max(1)));