This structure stores our comparator and allocator functions, as well
as a root node, ensuring we're using the same functions on one tree.
Makes the API much cleaner to use.
So the real purpose of arena_realloc is to figure out if we can get
away with just adjusting the region that the pointer given resides in
so that we don't need to _actually_ allocate any new memory.
The previous implementation did this in the special case where the
pointer given _is_ the entire region. But we can adjust the region's
size _if_ the pointer given is the last allocation on the region
i.e. it's on the tail end.
If someone uses the arena functionality, they may as well get the
region functionality for free. In particular, they may just want a
fixed bump allocator implementation which they can get here for free.
1) Better naming.
2) Use MIN macro for choosing the size to copy over from the old
buffer to the new one. Also fix issue where we free the old region of
memory before copying it over to the new buffer.
Manages individual allocations via a bump allocator (region), with
unfit allocations triggering a new region allocation that is added to
a linked list. Any new region allocations will always be oversized
for the initially requested size, to amortize the cost of future
allocations.
