Get fresh global statistics

When you access the statistics, it seems that you have two options: either print the "global" statistics with mi_stats_print_out or alternatively, each thread can print its own statistics.
Neither option is ideal for us, because we have a large pool of worker threads that never return (global statistics seem to be updated only when a thread exits) and each of them may allocate/free memory anytime; you can send a query to the process, but it will be served by a random thread in a different pool, so basically if you ask the stats, you get back a list of very small numbers, while the process itself is using gigabytes.

One option could be to enumerate all the stats objects of all heaps/threads and merge them on request into a temporary (without resetting them), but it's a bit unclear how to achieve this.

For the moment, we added a very crude grand total with a very simple patch: just to give you an idea, in alloc.c:

volatile _Atomic(uintptr_t) grandtotal = 0;

extern inline mi_decl_restrict void* mi_malloc(size_t size) mi_attr_noexcept {
mi_atomic_add(&grandtotal, (uintptr_t)size);
return mi_heap_malloc(mi_get_default_heap(), size);
}

extern inline mi_decl_restrict size_t mi_total_malloc_size() mi_attr_noexcept {
// not ideal, but it should work everywhere
return (size_t)mi_atomic_add(&grandtotal, 0);
}

and later in mi_free we subtract mi_page_usable_block_size(page)

however, I'd like to get an opinion on this, and I'm not totally sure of which entry points I should touch to update the grand total (e.g. mi_heap_malloc_small...).

(p.s.: there's a typo in the documentation:
[https://microsoft.github.io/mimalloc/group__extended.html#ga256cc6f13a142deabbadd954a217e228], where mi_stats_print_out is incorrectly called mi_stats_print)

Start in alloc.c with the existing grandtotal experiment, then trace mi_malloc, mi_free, and the mentioned mi_heap_malloc_small entry points to understand where statistics change. Review how mi_stats_print_out currently obtains global statistics and determine how a fresh aggregate could be produced without resetting counters. Check the extended documentation link for the mi_stats_print typo as a separate completion item.