Improve detect_change function

Improve detect_change function

Opening the valve to cell in two stages introduced a problem with
the peak detection (double peaks).
For small pressures during desorption measurement the pressure
differences were also small and the threshold could not be set so that
it recognizes these small peaks without triggering the double peaks.

Both seem to be somewhat solved by what I've done. The differences are
divided by the mean of differences 15 (arbitrary) sample to both sides.
This amplifies the peaks that occur around vacuum much more than others.
In other words the difference function is divided by the convolution of
differences and a rectangular window (of size 30).

The convolution looks like exponentials that roughly follow the
pressure that have smeared peaks. By dividing the differences by this
convolution, the small peaks are divided by small values while the big
peaks are divided by greater values. This maybe worsens the
noise-to-signal ratio, but makes the peaks easier to process.