Add casts inside matrix constructor

[Martin Sjölund]

The following code flattens wrongly:

model M
  function f
    input Integer i[:];
    input Real r[:];
    output Real o[:,:];
  algorithm
    o := [i;r];
  end f;
  Real mat[:,:] = f({1,2,3},{4,5,6});
end M;

It becomes:

  o := /*Real[:, 1]*/(cat(1, promote(i, 1), promote(r, 1)));

But should become:

  o := (cat(1, /*Real[:, 1]*/ promote(i, 1), promote(r, 1)));

[Martin Sjölund]

Fixed for this case. There might be more, but elabMatrix needs to be rewritten to handle everything since it seems to be done in a way that is very hard to make a small fix.