count bytes allocated through malloc more precisely

[d-netto]

Should make the accounting for memory allocated through malloc a bit more accurate.

Should also simplify the accounting code by eliminating the use of jl_gc_count_freed in jl_genericmemory_to_string.

Works fine on my M2, but let's wait for CI to see how portable this patch is.

[vtjnash]

SGTM. Assuming we can build on musl, I don't see a problem with this being de facto portable to any future memory allocator we would use

[gbaraldi]

I would just measure how expensive this call is on a hot path (I suspect it's super cheap so LGTM)

[d-netto]

Latest version is still not ready for merging though.

Even though the latest commits passed on Linux looks like the macOS-aarch64 tests hang (despite building fine on it).

[gbaraldi]

One thing that looking at this made me see is the malloc_good_size function that macos has. I wonder if we should take it into account for the overallocation things we have.

[d-netto]

On Windows:

C:/workdir/src/gc-common.c:122:12: error: implicit declaration of function '_aligned_msize'; did you mean '_aligned_free'? [-Wimplicit-function-declaration]
  122 |     return _aligned_msize(p, JL_CACHE_BYTE_ALIGNMENT, 0);
      |            ^~~~~~~~~~~~~~
      |            _aligned_free

Microsoft docs tell that this function should be in the <malloc.h> header: https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/aligned-msize?view=msvc-170, so the failure seems confusing to me.

[d-netto]

Not pursuing it further.