test: widen litest to use doubles for scaled variables

test: widen litest to use doubles for scaled variables

Using a 0-100% range is useful but in some cases we need events with finer
than 1% granularity.

And fix up the two-finger test that now fails. This was a bug in the test
anyway, the dx/dy supplied here was 1% of the touchpad width. Confined to
integers this meant we only ever had the touch down, then the single move by
1%. That caused two events - not enough to satisfy tp_estimate_delta, so we
always had a delta of 0/0 regardless of the size of the move.

Now with doubles this fails, so drop it to 0.1% instead, which is small enough
on all touchpads we currently have.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>