Implement a lock-free malloc/free for local allocations

[catamorphism]

instead of relying on the standard malloc/free, as per FIXME in rt/memory_region.h

[bblum]

We now have jemalloc, the specifics of which I'm not familiar with. Is it good enough?

[bblum]

@brson @pcwalton

[thestinger]

jemalloc is lock-free for allocations under 4K, but not above that

I don't think the per-task heap will be using malloc/free when @graydon's gc lands anyway

[brson]

jemalloc is turned off again. I don't think graydon's gc implements a custom local allocator, which is what this issue is about, but it'll probably be hard to beat jemalloc even for local allocations. Updated title to reflect that this is for the local allocator.

[brson]

Dupe of #3472