This makes it possible to include all .cc files into build, even if not
building CoreText, Uniscribe, etc.

This was mostly to help custom builders.  But also means that we can
include all files in our own build system.  Not sure if we should.
Definitely simplifies things, but slightly only.

This reverts commit 47030b18.

Reverts only the hb-ft part, to reinstate non-OpenType kerning with
FreeType.

Fixes https://github.com/harfbuzz/harfbuzz/issues/1682

If compiler doesn't inline StructAtOffset, this was an error since we
only disable cast-align at call-site.  So, move the cast out.

../src/hb-machinery.hh: In instantiation of 'const Type& StructAtOffset(const void*, unsigned int) [with Type = unsigned int]':
../src/hb-font.cc:146:85:   required from here
../src/hb-machinery.hh:63:12: error: cast from 'const char*' to 'const unsigned int*' increases required alignment of target type [-Werror=cast-align]
 { return * reinterpret_cast<const Type*> ((const char *) P + offset); }
            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../src/hb-machinery.hh: In instantiation of 'Type& StructAtOffset(void*, unsigned int) [with Type = unsigned int]':
../src/hb-font.cc:147:79:   required from here
../src/hb-machinery.hh:66:12: error: cast from 'char*' to 'unsigned int*' increases required alignment of target type [-Werror=cast-align]
 { return * reinterpret_cast<Type*> ((char *) P + offset); }
            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

That's better use of that value than requiring extra macro HB_NO_ATEXIT

* Use non-GRID-fitted values for metrics

See freetype/src/base/ftobjs.c:ft_recompute_scaled_metrics() and
the usage of GRID_FIT_METRICS inside.

Fixes https://github.com/behdad/harfbuzz/issues/1262

* Update hb-ft.cc

Saves 4 or 8 bytes per object on 64bit archs.

Fixes https://github.com/harfbuzz/harfbuzz/issues/1299

Removes anomaly I was seeing in cpal table trying to use implicit Null(NameID).

Makes our FT-backed hb_font_t safe to use from multiple threads.  Still,
the underlying FT_Face should NOT be used from other threads by client
or other libraries.

Maybe I add a lock()/unlock() public API ala PangoFT2 and cairo-ft.
Maybe not.

Some more measurable speedup.  The recent commits' speedups are as follows:

Testing with Roboto, ****when disabling kern and liga****:

Before:

FT --features=-kern,-liga
user↦   0m0.521s

OT --features=-liga,-kern
user↦   0m0.568s

After:

FT --features=-liga,-kern
user↦   0m0.428s

OT --features=-liga,-kern
user↦   0m0.470s

So, 17% speedup.

Note that FT callbacks are faster than OT these days since we added an advance
cache to FT.  I don't think the difference is enough to justify adding a cache
to OT.

When not disabling kern, the thing is three times slower, so the speedups
are three times less impressive...  Still, 5% not bad for a codebase that I
otherwise thought is optimized out.

Note that, because of this and other optimiztions in our main shaper,
disabling kern and liga, the OT shaper is now *faster* than the fallback
shaper.  So, that's my recommendation to clients that need the absolute
fastest...

Ouch!

And remove redundant implementations.

I decided to always use the cache, instead of my previous sketch direction
that was to only allocate and use cache if fast advances are not available.
The cache is a mere 1kb, so just use it...

TODO: Invalidate cache on font size change.

Fixes https://github.com/harfbuzz/harfbuzz/issues/651
Fixes https://github.com/harfbuzz/harfbuzz/pull/1082

Related to https://github.com/harfbuzz/harfbuzz/pull/1082

This reverts commit 473b17af.

Updates to recent changes.

Sorry for the noise, downstream custom builders.  Please adjust.

Add HB_USE_ATEXIT macros around free_static function definitions to
avoid unused function compiler warnings/errors.

Found and fixed a couple bugs.

Found a couple multithreading issues.  Marked them with "XXX-MT-bug".