Randomization only happens by default. If the user specifies a value for
'rand', that value is respected.

Fixes https://github.com/harfbuzz/harfbuzz/issues/1141

New API:
HB_FEATURE_GLOBAL_START
HB_FEATURE_GLOBAL_END

Fix a bug in CBDT sanitize, and redundant check in avar.

Fix avar sanitize().

Fixes https://github.com/harfbuzz/harfbuzz/issues/1154

Fixes https://github.com/harfbuzz/harfbuzz/issues/556

Devanagari regresses 12 tests, and Gujarati 2.  See:

  https://github.com/harfbuzz/harfbuzz/issues/556#issuecomment-419957472

New numbers:

BENGALI: 353725 out of 354188 tests passed. 463 failed (0.130722%)
DEVANAGARI: 707299 out of 707394 tests passed. 95 failed (0.0134296%)
GUJARATI: 366353 out of 366457 tests passed. 104 failed (0.0283799%)
GURMUKHI: 60729 out of 60747 tests passed. 18 failed (0.0296311%)
KANNADA: 951300 out of 951913 tests passed. 613 failed (0.0643966%)
MALAYALAM: 1048136 out of 1048334 tests passed. 198 failed (0.0188871%)
ORIYA: 42327 out of 42329 tests passed. 2 failed (0.00472489%)
SINHALA: 271596 out of 271847 tests passed. 251 failed (0.0923313%)
TAMIL: 1091754 out of 1091754 tests passed. 0 failed (0%)
TELUGU: 970555 out of 970573 tests passed. 18 failed (0.00185457%)

Fixes https://github.com/harfbuzz/harfbuzz/issues/1110

This reverts commit 616fd34a.

So I can get a release out.  I haven't debugged those yet, and they
are producing bad tables.

John Emmas reported on mailing list that build with MSVC is failing.
Not sure why bots don't catch it, but this should fix.

Technically, static const needs an out-of-class definition.  Eg:

  CXXLD    libharfbuzz-subset.la
Undefined symbols for architecture x86_64:
  "OT::FeatureVariationRecord::min_size", referenced from:
      bool OT::GSUBGPOS::subset<OT::PosLookup>(hb_subset_context_t*) constin libharfbuzz_subset_la-hb-subset.o
      bool OT::GSUBGPOS::subset<OT::SubstLookup>(hb_subset_context_t*) constin libharfbuzz_subset_la-hb-subset.o
  "OT::Record<OT::LangSys>::min_size", referenced from:
      OT::Script::subset(hb_subset_context_t*) constin libharfbuzz_subset_la-hb-subset.o
  "OT::IntType<unsigned short, 2u>::min_size", referenced from:
      OT::Script::subset(hb_subset_context_t*) constin libharfbuzz_subset_la-hb-subset.o
      OT::RecordListOf<OT::Feature>::subset(hb_subset_context_t*) const  in libharfbuzz_subset_la-hb-subset.o
ld: symbol(s) not found for architecture x86_64
collect2: ld returned 1 exit status
make[4]: *** [libharfbuzz-subset.la] Error 1
make[3]: *** [all-recursive] Error 1
make[2]: *** [all] Error 2
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2
Exited with code 2

They are still in the drop list.

Should remove the vector version at some point...

Why does subset plan not have a hb_set_t of glyphs?

Ouch.

Remove Lookup::subset().

Add subset() call for GSUBGPOS struct and its dependencies.
Not hooked up anywhere.

We're going to (finally) use virtual methods for hb_serialize_context_t
customization, so don't need to carry a Serializer template variable
around...  Simplifies code.