On some machines (nami), interrupt latency cause samples to appear
to be from the future and are pegged to the current time.
We would see samples with this pattern:

[t, t + ~5ms, t + ~10ms, t + ~10ms + 100us, t + ~10ms + 200us],
                             (current now)      (current now)
(t is the last timestamp time)

Last 2 samples would be barely spread, causing applications to
complain.

We now spread the entire sequence. This is not great: in the example
the sensor was supposed to send samples every 5ms, it now appears to
send one every 2.5ms, but it is slightly closer to reality:

sampling time in the example above
At sensor level

1             2             3             4            5
+-----5ms-----+-----5ms-----+-----5ms-----+----5ms-----+---> t

Before, at host level
                       1             2             3 4 5
--interrupt delay------+-----5ms-----+-----5ms-----+-+-+---> t

Afer, at host level
                       1       2       3       4       5
--interrupt delay------+-2.5ms-+-2.5ms-+-2.5ms-+-2.5ms-+---> t
Signed-off-by: Gwendal Grignou <gwendal@chromium.org>
Signed-off-by: Enric Balletbo i Serra <enric.balletbo@collabora.com>