On-the-fly interpolation of collision coeffs.

[tlunttil]

Implementation of the on-the-fly interpolation of collisional
coefficients that was discussed in comments to issue #81. With the
default example/model.c memory saving is about 50% and there is no
effect on speed.

[allegroLeiden]

This all looks good, I just have a few stylistic quibbles as follows:

• molinit.c, line 271: the loop could and should be omitted, together with the commented-out lines within (also other commented-out lines in molinit()).

• molinit.c, lines 290-295: this 'else' block (inherited from the previous code) could be made a small bit more conservative/safer by replacing it with something like

    } else if(g[id].t[0]<=part[ipart].temp[0]) {
      g[id].mol[i].partner[ipart].t_binlow=0;
      g[id].mol[i].partner[ipart].interp_coeff=0.0;
    } else {
      g[id].mol[i].partner[ipart].t_binlow=ntemp[ipart]-2;
      g[id].mol[i].partner[ipart].interp_coeff=1.0;
    }
  

• stateq.c, line 128 (applies also to line 116 as inherited from the previous code): personally I would not use 't' here, since it makes one think of temperature. 'ti' perhaps, with 'li' at line 116.

• stateq.c, lines 129-131: instead of 'trans_offset' defined as given, why not just something like

  int i = t*m[ispec].ntemp[ipart]+g[id].mol[ispec].partner[ipart].t_binlow;

Then the next 2 lines (actually a single line, wrapped) would become much more compact/readable as well as more conservatively expressed:

  double down = downrates[i] + g[id].mol[ispec].partner[ipart].interp_coeff*(downrates[i+1] - downrates[i]);

[tlunttil]

Yep, I forgot to remove the commented-out old parts. I'll do that, and your other suggested changes, and push a new version probably later today.