gc_alloc_var optimization

[ncannasse]

ATM gc_alloc_var takes too much time after a mark to allocate on already filled pages, because it will not blacklist full pages and will keep iterating over them. Using gc_free_pages as we do for gc_alloc_fixed would help.

This can be reproduced by disable interface memoization and running Northgard main loop.

[ncannasse]

I did some fixes today, allocation is still a bit expensive compared to some other GC techniques, I'll have to review the GC design later.