Rework the interrupt handling to avoid caching the values as the core is
already doing that. The core also always ensures the rtc_time passed for
the alarm is fully populated.

The only trick is in read_alarm where status needs to be read before the
alarm registers to ensure the potential irq is not cleared.

Link: https://lore.kernel.org/r/20191019204941.6203-8-alexandre.belloni@bootlin.comSigned-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>