stable-qsort: avoid using potentially unaligned access

Like in the previous patch for compat/qsort_s.c, remove the optimization
of using an on-stack buffer to avoid small allocations. This ensures
maximum alignment for the array elements and simplifies the code a bit.

The performance impact for the current callers is unlikely to be
noticeable:

* compat/mingw.c::make_environment_block() uses ALLOC_ARRAY and
ALLOC_GROW several times already, so another allocation of up to 1KB
should not matter much.

* diffcore-rename.c::diffcore_rename_extended() is called once per diff
or twice per merge, and those require allocations for each object and
more already.

* merge-ort.c::detect_and_process_renames() is called once per merge.
It's responsible for the two per-merge diffcore_rename_extended()
calls mentioned above as well, though. So this is possibly the most
impacted caller. Per-object allocations are likely to dwarf the
additional small allocations in git_stable_qsort(), though.

Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>