Misc memory fixes

[mtzguido]

Most importantly this one 57a40dd. Seems to be a slight improvement in the graphs. But, the memory usage graph seems to be too noisy... there's no way this significantly increases memory usage. My only guess is fragmentation. I can make it more deterministic by tuning OCaml's GC with max_overhead=0 to trigger compaction on every GC, thought that makes things significantly slower. Maybe memory profiles should run with that setting?

guido@GuidoX1:~/FStar(master)$ .scripts/runlim_diff.py --dirs . ../pristine/
Comparing . and ../pristine/, --dirs=True
F* runlim time slope: 0.9859187509074009
F* runlim space slope: 0.9284346462377224