Muselet memory leak

[Tenivar]

Source Extractor : 2.25, from conda forge
numpy 1.17.3
Python 3.7.6
Astropy 4.0.1
Ubuntu 18.04.3 LTS
mpdaf 3.4

When applying Muselet on "huge" datacubes, on one or several cpus, some of the step 1 variables are not collected by the garbage collector. If Muselet run successively on various datacubes, that causes a huge consumption of RAM that is only freed when the Python kernel is restarted.