to_gdx seems to take a lot of RAM

[jebob]

In my test case, I am writing some large parameters (1x 5 dim x 29M elements, 1x7dimx29M elements, 3 small <200k element parameters) to a gdx using to_gdx. This takes somewhere on the region of 18GB of RAM, despite the resulting GDX being only 0.4GB.