Works beautifully!  Test coming.

Assumes graphemes only have one base glyph.

Untested.

This concludes kerx table support!

Untested.

Tested with Kannada MN:

$ HB_OPTIONS=aat ./hb-shape Kannada\ MN.ttc -u 0CCD,0C95,0CD6
[kn_ka.vattu=0+230|kn_ai_length_mark=1@326,0+607]

Doesn't apply actions yet.

Tested using Kannada MN:

$ HB_OPTIONS=aat ./hb-shape Kannada\ MN.ttc -u 0C95,0CCd,C95,CCD
[kn_ka.virama=0+1299|kn_ka.vattu=0+115|_blank=0@-115,0+385]

$ HB_OPTIONS=aat ./hb-shape Kannada\ MN.ttc -u 0C95,0CCd,C95,CCD --features=-kern
[kn_ka.virama=0+1799|kn_ka.vattu=0+230|_blank=0+0]

I don't see the GPOS table in the font do the same.  ¯\_(ツ)_/¯

Untested.

Untested.  The only Apple font shipping with this format is San Francisco fonts
that use this for their kerx variation tables, which we don't support.

Tested that Format0 works with Kannada MN font:

$ make -j5 lib -s && HB_OPTIONS=aat ./hb-shape Kannada\ MN.ttc -u 0C95,0CC2
[kn_ka=0+1000|kn_matra_uu=0@-30,0+1345]

$ make -j5 lib -s && HB_OPTIONS=aat ./hb-shape Kannada\ MN.ttc -u 0C95,0CC2 --features=-kern
[kn_ka=0+1030|kn_matra_uu=0+1375]

Note that GPOS does the same with 'dist' feature, and applies the whole difference to the
same glyph:

$ make -j5 lib -s && ./hb-shape Kannada\ MN.ttc -u 0C95,0CC2
[kn_ka=0+970|kn_matra_uu=0+1375]

$ make -j5 lib -s && ./hb-shape Kannada\ MN.ttc -u 0C95,0CC2 --features=-dist
[kn_ka=0+1030|kn_matra_uu=0+1375]

Disable kern Format2.

Fix kerx Format2.  Manually tested this with Tamil MN font and it works:

$ HB_OPTIONS=aat ./hb-shape Tamil\ MN.ttc -u 0B94,0B95
[tgv_au=0+3435|tgc_ka=1@-75,0+1517]

 HB_OPTIONS=aat ./hb-shape Tamil\ MN.ttc -u 0B94,0B95 --features=-kern
[tgv_au=0+3510|tgc_ka=1+1592]

Fixes https://github.com/harfbuzz/harfbuzz/issues/322

We never tested compiling without it.  Just kill it.  We always build
our own shaper.

This speeds up Roboto shaping by ~10%.  I was hoping for more.
Still, good defense against lookups with many subtables.

In preparation to move add per-subtable set digests...