Using font from https://github.com/behdad/harfbuzz/issues/300

Fixes https://github.com/behdad/harfbuzz/issues/300

Limits number of alternatives per glyph per feature to 255, so be it.
That's better than possibly breaking shaping because of one bad feature
value.

This one:

  map->mask = (1 << (next_bit + bits_needed)) - (1 << next_bit);

before the fix, the shift was done as an int, causing overflow
if it ever got to 1 << 31.  Sprinkle 'u's around.

Fixes https://bugs.chromium.org/p/chromium/issues/detail?id=634805

Fixes https://github.com/behdad/harfbuzz/issues/300

Regression from: b424b6c3.

Fixes https://github.com/behdad/harfbuzz/issues/297

Fixes https://github.com/grigorig/ucdn/issues/12

This is based on bug reports that have been filed against Firefox since it
updated to a version of harfbuzz that uses zeroing by GDEF rather than by
Unicode. I'm sure there are a bunch more font versions that should also be
included; these are just the ones I have on hand and have confirmed as having
bad GDEF data.

Given how the list here is growing, I think we should reconsider the approach,
and perhaps revert to zeroing by Unicode instead.

Fixes https://github.com/behdad/harfbuzz/issues/264
Fixes https://github.com/behdad/harfbuzz/pull/266

Fixes https://github.com/behdad/harfbuzz/issues/236
Fixes https://bugs.chromium.org/p/chromium/issues/detail?id=627953

Fixes https://bugs.chromium.org/p/chromium/issues/detail?id=614647
If allocation fails, we might be leaving junk behind.  At least
clear it up.

Prior to this change the function `FT_Error FT_Done_Face(FT_Face *)` was
called through a pointer with the signature `void (void *)` resulting in
undefined behaviour.

[introspection] Fix hb_tag_from_string annotation

Fixes https://github.com/behdad/harfbuzz/pull/271

Discard reconfigured CTFont if URL changes

Part of https://github.com/behdad/harfbuzz/issues/277

Revert "use utf32"

	[dwrite] Use stream font loader instead GDI interop

Addressing Nikolay Sivov reviews on harfbuzz mailing list

No longer noticeable memory increase on create/destroy iterations,
highly better than current state of uniscribe backend

This reduces memory consumption of my iterated font create/destroy cycle test
significantly and makes it much better than uniscribe backend even

With help of
https://dxr.mozilla.org/mozilla-central/source/gfx/2d/NativeFontResourceDWrite.cpp

Fix make check on Travis

The so-called Python 2 “narrow” builds support UCS2 only, this is a
workaround to allow unichr to work with any Unicode character in such
builds. This fixes Travis-CI failure as it has narrow Python 2 builds.

Copied from:
https://github.com/behdad/fonttools/blob/master/Lib/fontTools/misc/py23.py

Fixes https://github.com/behdad/harfbuzz/issues/267

Fixes https://github.com/behdad/harfbuzz/issues/255