This also factorize the loading codepaths with and without slicing

Just checking boolean value would lead to a len() call, which might fail
as line_idxs might not have been set yet

self.layers dictionnary could change from another thread, thus changing
size during repaint iteration. Using a fixed range with the current
number of layers avoids iterating directly on the dictionnary.

Now all we need is to find how to handle the numpy.fromiter calls in the
3D viewer to have the view actually show something before the end of the
loading.

This way we can easily yield at the end of each iteration. The remaining
issue is mostly how to handle the costly numpy.fromiter, which we don't
want to do uselessly several times.

Also fix the wrong behavior of estimate_duration on new calls

This made them fail when data was being written on standard outputs

Any host command (other than pause) will now be run as a pronsole
command. This will allow doing things such as #428 easily.