We still only use the first subtable, but we no longer fail if there's
more than one subtable.

Updates fogleman/gg#7

For example, "Apple Color Emoji.ttc" has such a cmap. After this commit,
we can render its glyph for "U+0001F1E6 REGIONAL INDICATOR SYMBOL LETTER
A", which is clearly outside the Unicode BMP.

Note that this only lets us render black-and-white glyphs for out-of-BMP
runes. Rendering color glyphs (i.e. emoji) remains TODO.

Updates #45

Support TTC v2

Close #11

This goes against spec, but matches the C Freetype implementation and
works around a bug in DejaVuSansMono.ttf's hinted '2' glyph.

The golang-dev discussion is at
https://groups.google.com/d/topic/golang-dev/qKpxZiNABcY/discussion

A recent change added a Metrics method to the font.Face interface.

Fix Capper & Joiner bugs introduced by 856a70c3.

Some constants were not updated properly when switching
from raster.Point to fixed.26_6

Fix max alpha value: 1<<32-1, not 1<<16-1.

This was missed in 8a442816.

Looks like since https://github.com/golang/freetype/commit/8a4428169df5c9ec333fbda61dca6b830cb47d66 you forget to change this in MonochromePainter

See the discussion on https://github.com/golang/freetype/pull/19

benchmark                 old ns/op     new ns/op     delta
BenchmarkDrawString-8     8618239       8149527       -5.44%

This tracks an upstream change in the x/exp/shiny/font package.

This is more consistent with the standard library's image/color package.

The benchmarks didn't seem to change significantly.

The slice Foos is plural; it refers to multiple elements. The element Foos[i]
is the i'th Foo.

The previous "the endpoints are inclusive" comment seems confusing. It's true
that the bounding box's max X equals the right-most coordinate, which suggests
<= instead of <, but that node's coordinate is itself exclusive. Consider the
solid 1-pixel square: (0, 0), (64, 0), (64, 64), (0, 64) in fixed.Point26_6
coordinates. The right-most coordinate is 64, and the bounding box's max X
equals 64, but rasterizing that square only affects sub-pixels up to but not
including 64.

Instead, it seems accurate to follow the fixed.Rectangle26_6 description, in
that the max values are exclusive.

The zero value of a GlyphBuf is already perfectly usable.

benchmark                 old ns/op     new ns/op     delta
BenchmarkDrawString-4     88103151      14188817      -83.90%

This is in anticipation of caching glyph images. Quantization means that cache
hits are more likely.

Also make NewFace take an *Options instead of an Options.

benchmark                 old ns/op     new ns/op     delta
BenchmarkDrawString-4     96914389      88683659      -8.49%

benchmark                 old allocs     new allocs     delta
BenchmarkDrawString-4     32287          0              -100.00%

benchmark                 old bytes     new bytes     delta
BenchmarkDrawString-4     1929619       156           -99.99%

These comments weren't updated when we went from 24.8 to 26.6.

Its implementation is mostly a copy/paste of the freetype.Context type.
Follow-up commits will make it more efficient.

Also add an example that uses a truetype.Face and the
golang.org/x/exp/shiny/font package to draw text.