Difference between mimalloc's freed report and valgrind --leak-check=full

Nobody has claimed this yet.

I have been investigating whether we should use mimalloc for [MoarVM](https://github.com/MoarVM/MoarVM) (the primary backend for the [Rakudo](https://github.com/rakudo/rakudo) Raku compiler) and initial results are very promising. However, am wondering about a difference I see between the stats reported when I use `MIMALLOC_SHOW_STATS=1` with `LD_PRELOAD=/path/to/libmimalloc-debug.so` compared to `valgrind --leak-check=full`. For a minimal test case `MIMALLOC_SHOW_STATS=1 LD_PRELOAD=/path/to/libmimalloc-debug.so raku --full-cleanup -e ''` (where the `--full-cleanup` flag to `raku` tells the VM to free every thing on exit instead of just letting the OS cleanup), mimalloc reports 1.2 MiB total for `current`. However, for `valgrind --leak-check=full raku --full-cleanup -e ''`, valgrind reports `All heap blocks were freed -- no leaks are possible`. Is this just an artifact of using dynamic loading? Or could there be some other reason?

Reproduce the minimal Raku command with MIMALLOC_SHOW_STATS and the Valgrind leak-check command, then compare what each tool counts as live memory. No source files or tests are named; done means explaining the 1.2 MiB discrepancy or identifying a concrete allocator accounting issue.

Written by the indexing model from the issue text.