fpc1020: Use atomic_{read,set} instead to avoid race

 * As seen in tissot's fpc1020 driver.
 * This is generally a more popular approach to avoid race when multiple
   threads may write and read a variable at the same time.