[pctablet505]

Fixes #739

Under uvloop, asyncio.sleep(n) can return up to ~1 ms before n has actually elapsed. This change pads the timeout passed to libuv by 1 ms, so timers never fire early.

Problem

libuv refreshes loop->time as the monotonic clock floored to whole milliseconds, and uv_timer_start() schedules the callback at loop->time + timeout. Whatever fraction of a millisecond had already passed when the clock was sampled is lost, so the timer comes due early in real time. The issue shows asyncio.sleep(0.1) returning after 0.0991 s.

Example with a 100 ms timer started at real time 1000.9 ms:

	loop->time at start	due at	real time slept
before	1000 (floored)	1000 + 100 = 1100	99.1 ms ❌
after	1000 (floored)	1000 + 100 + 1 = 1101	100.1 ms ✅

Fix

Only the value handed to libuv changes (self.timeout + 1). self.timeout, and therefore TimerHandle.when() / get_when(), still reports the requested deadline. The trade-off is that a timer can fire up to ~1 ms later than before, which is within libuv's 1 ms resolution.

This is a different truncation from #233, which was fixed by rounding the float-to-ms conversion in call_at(). That path is unchanged.

Tests

test_call_later_not_early runs asyncio.sleep(0.01) 300 times and asserts that each one lasts at least the delay, measured with time.monotonic(). The existing test_call_later_rounding (#233) can't catch this, because it measures with the loop's own millisecond-floored clock.