Bugfix: Closes #15

[HealthyPear]

In Issue #15 it has been reported that the new ctapipe-based cleaning for the biggest cluster (or island) was crashing in case of no islands surviving the cleaning algorithm.

This has been now fixed by checking beforehand the number of surviving islands and calling the function largest island only in case of more than 1 island left in the image.
In the case of only 1 island left, the cleaning mask is the original one.
In the case of no islands, the original (empty) cleaned image is used.
A similar check is applied in the case of extended cleaning, which would take into account all the surviving islands but which for the moment is set to be the same as the biggest cleaning.

The new code has been tested successfully on 500 events from file gamma_20deg_180deg_run100___cta-prod3-demo-2147m-LaPalma-baseline.simtel.gz using both DL1 and DL2 scripts (the latter without model estimation, so the result is the same as for DL1):

EventCutFlow
   Cut Name   selected Events Efficiency
------------- --------------- ----------
       noCuts             500      1.000
min2Tels trig             500      1.000
min2Tels reco             349      0.698
direction nan             349      0.698

ImageCutFlow
     Cut Name     selected Events Efficiency
----------------- --------------- ----------
           noCuts            2313      1.000
        min pixel            2157      0.933
       min charge            2119      0.916
     poor moments            2119      0.916
  bad ellipticity            1578      0.682
close to the edge            1966      0.850

After this PR the new bugfix version 0.2.1 will be released.