Added in 01dff1ea and 19256bef, this was targeted at older
msvc versions that don't support C99 but now as we require
C++11 we don't target places those envs thus removing this.

+this works on gcc 4.8 as well as default code path.

This reverts commit a0c49007.

Just testing bots.  Will finish based on results.

__OPTIMIZE_SIZE__ should be defined whenever -Os but some Internet thread
indicate may not so lets do that ourselves as that is the main intention

The only thing I need for a working wasm in a minimum libc,
otherwise I have to provide the very same qsort inside that libc

We are not testing Windows exes in fedora mingw bot, we don't have to as
probably won't go that smoothly and we have real Windows bots anyway
and as wine installation itself is time taking let's remove it

https://github.com/harfbuzz/harfbuzz/issues/1652

Fixes https://github.com/harfbuzz/harfbuzz/issues/1724

https://github.com/harfbuzz/harfbuzz/issues/1740

I guess all of its field will be initialized anyway here but lets make it more defensive

Causing -Wextra-semi-stmt build error when no primitive has chosen
Interesting that nobody has noticed it yet.

mingw.org's MinGW.

src/hb-algs.hh: Don't compile _BitScanForward and _BitScanReverse
for GCC >= 4.  mingw.org's MinGW doesn't have these functions.

src/hb-atomic.hh: MemoryBarrier does exist in mingw.org's MinGW,
but it is not a macro, it is an inline function.  __MINGW32_VERSION
is a macro that exists only in mingw.org's MinGW, so conditioning
on it should not affect MinGW64, where MemoryBarrier is a macro.

src/hb-uniscribe.cc: Define E_NOT_SUFFICIENT_BUFFER if it is not
defined (mingw.org's MinGW doesn't).

src/hb.hh: Don't include intrin.h for mingw.org's MinGW, since that
header is not available; instead, include windows.h.  Conditioned
on __MINGW32_VERSION to avoid affecting MinGW64.

Saves a few bytes.

Don't use vsnprintf when HB_NO_BUFFER_MESSAGE is defined

Sometime between [July 2018] and [January 2019], the OpenType Myanmar
grammar changed: one asat is now allowed after a medial ya, before any
other medial consonant.

[July 2018]: http://web.archive.org/web/20180711011550/https://docs.microsoft.com/en-us/typography/script-development/myanmar
[January 2019]: http://web.archive.org/web/20190115044451/https://docs.microsoft.com/en-us/typography/script-development/myanmar

This also reverts commit 439b0586, which
allowed an asat immediately after a medial ra.

fixed faulty FDSelect::sanitize

Originally introduced in 21ede867 (#1314) but as it wasn't applied to hb_bsearch
accidentally removed while merging hb_bsearch_r to it.

It should've done in a branch