port umm_malloc as new allocator

[bhansconnect]

This is a direct port of the umm_malloc a version of malloc meant for embedded systems. By having a proper free list, we are able to track allocations much faster and avoid the performance bottlenecks that come from scanning bitset.

On top of that, this allocator will allocate to the space of best fit instead of the first location with enough space (this reduces fragmentation and memory pressure). Lastly, the allocator has a proper realloc algorithm that avoids a lot of problems with the naive implemention of malloc, copy, free. Overall, it should be a large perf and memory usage win.

It does add 4 bytes of overhead to each allocation. The old allocator had 1 bit of overhead for every 8 bytes allocated. So there is definitely more overhead. That said, the fragmentation and perf gains outweigh the costs in my opinion. Also, for wasm4, really only lists and large strings are allocated. Those tend to be large enough to pay for the overhead.

[lukewilliamboswell]

Thank you @bhansconnect 😍