Add a flag to hb_buffer_t to prevent the insertion of dotted circles on incorrect character sequences. 

Current behavior unchanged if this flag is not set (and it isn't by default).

I stumbled over this when trying to upgrade the version of HarfBuzz used by
LibreOffice to 3.2.1 (see <https://gerrit.libreoffice.org/plugins/gitiles/core/
+/b7ddc514bff9bdf682abae537f990aa01dc2c0fb%5E!/> "Upgrade to latest
HarfBuzz 2.3.1"), where building with MSVC 2017 failed like

> c:\cygwin\home\tdf\lode\jenkins\workspace\gerrit_windows\workdir\unpackedtarball\harfbuzz\src\hb-atomic.hh(272): error C2440: 'reinterpret_cast': cannot convert from 'const int *' to 'std::atomic<int> *'
> c:\cygwin\home\tdf\lode\jenkins\workspace\gerrit_windows\workdir\unpackedtarball\harfbuzz\src\hb-atomic.hh(272): note: Conversion loses qualifiers
> c:\cygwin\home\tdf\lode\jenkins\workspace\gerrit_windows\workdir\unpackedtarball\harfbuzz\src\hb-atomic.hh(272): error C2227: left of '->load' must point to class/struct/union/generic type

(see <https://ci.libreoffice.org/job/gerrit_windows/29916/>).

I added all the necessary "const" to make building of HarfBuzz 2.3.1 with
MSVC 2017 succeed for me.  There may be more missing at least conceptually.

As vscode is going to support it soon

It makes a hb_face_t from IDWriteFontFace, useful when using
DirectWrite facilities for font selection, loading and rendering
but using harfbuzz for shaping.

Fixes https://github.com/harfbuzz/harfbuzz/issues/1596

in case called immediately after init()

It is rather confusing to have script, language etc, in hb-common section
while feature is in hb-shape section. I keep looking for it in hb-common
section then using the API index because I can’t find it there.

Fixed a small typo.
[skip ci]

Copied and edited from the util option documentation. The docbook table
syntax is too verbose, but that is the best I can come up with.

CFF more arrayof fixes

[subset] Implement --retain-gids with CFF/CFF2

Wish there was a way to streamline this :(.

For some reasons djgpp doesn't understand "unsigned int" can be same
with one of uint*_t anyway so lets do that for it explicitly.

Just to note, our CI's djgpp is based GCC 7.2.0 and isn't old.