[SWIG] split stir.i and change order of declarations

[KrisThielemans]

The GeneralisedObjectiveFunction hierarchy was defined before normalisation and projectors, leading to problems setting/getting corresponding members.

As this was very messy, I've split stir.i in several files. It still leads to one huge *wrap.cxx but at least reading stir.i is now a bit easier. (still work to do).

This commit also fixes a line in SWIG for ProjData::get_subset, but that function is still ignored, so it doesn't change the actual Python interface.

Fixes #1018

[KrisThielemans]

@markus-jehl can you please check this? Also, because of the reordering, the FBP problem tha tyou had might be resolved, so you could try to enable the "instantiations" (now in stir_reconstructors.i)

[markus-jehl]

Sure, I'll have a look if this still works with my reconstruction pipeline and if that solves the FBP problem! Will probably be next week, though.

[robbietuk]

I like the simplification of stir.i. I was able to compile this PR and I now have access to stir.BinNormalisationFromProjData(), which addresses #1018

[markus-jehl]

I just checked it and it all works fine for me as well. Defining the FBP templates no longer crashes everything, so that's a plus in my view :-) I still can't define the FBP templates in a way that exposes their member functions, but this is not just about working out how this needs to be done in SWIG.

[KrisThielemans]

thanks. can you create a new issue for the FBP problem (maybe there's one already..)