cairo: report the cap height and measure the x-height

[probonopd]

On the cairo backend -[NSFont capHeight] is 0 for every font, and -xHeight is a fixed 60 percent of the ascender rather than the font's own value. Anything that centres a line of text on one of these metrics, a table cell or a custom control drawing its own label, then sits several points too high. Seen with a hand-drawn source list cell whose labels were visibly above the row centre.

Reproducer, a small tool printing the metrics of three fonts:

#import <AppKit/AppKit.h>
int main(void)
{
  @autoreleasepool
    {
      [NSApplication sharedApplication];
      for (NSFont *f in @[[NSFont systemFontOfSize: 13],
                          [NSFont boldSystemFontOfSize: 11],
                          [NSFont userFontOfSize: 12]])
        printf("%s %g: ascender %.2f descender %.2f capHeight %.2f xHeight %.2f\n",
               [[f fontName] UTF8String], [f pointSize], [f ascender],
               [f descender], [f capHeight], [f xHeight]);
    }
  return 0;
}

Before (master, Inter as the system font):

Inter-Medium 13: ascender 12.59 descender -3.14 capHeight 0.00 xHeight 7.56
Inter-Bold 11: ascender 10.66 descender -2.65 capHeight 0.00 xHeight 6.39
Inter-Medium 12: ascender 11.62 descender -2.89 capHeight 0.00 xHeight 6.97

After:

Inter-Medium 13: ascender 12.59 descender -3.14 capHeight 9.46 xHeight 7.10
Inter-Bold 11: ascender 10.66 descender -2.65 capHeight 8.00 xHeight 6.00
Inter-Medium 12: ascender 11.62 descender -2.89 capHeight 8.73 xHeight 6.55

Inter's cap height is 0.727 em, so 9.46 at 13 points is the real value, and its x-height is 0.546 em, so the old estimate was 6 percent high. The change takes both from cairo_scaled_font_text_extents of "H" and "x" on the scaled font, which works for every face cairo renders, and keeps the old estimate for the x-height only when the face has no "x" glyph.

The second commit fixes -boundingRectForGlyph: on the same backend, which returned cairo's y-down extents unchanged: y_bearing is the top of the ink measured downwards from the origin, so the rect sat below the baseline in the font's y-up coordinates, unlike the Xlib and Xft backends, which return the origin at the bottom of the ink. Same tool, printing the rect of four glyphs of the 13 point system font, before:

H: origin (1.05, -9.46) size (7.59 x 9.46)
x: origin (0.36, -7.10) size (6.52 x 7.10)
g: origin (0.61, -7.19) size (6.51 x 9.99)
p: origin (0.94, -7.19) size (6.49 x 9.84)

After:

H: origin (1.05, 0.00) size (7.59 x 9.46)
x: origin (0.36, 0.00) size (6.52 x 7.10)
g: origin (0.61, -2.81) size (6.51 x 9.99)
p: origin (0.94, -2.65) size (6.49 x 9.84)

The origin is now -(y_bearing + height): the bottom of the ink, 2.81 points below the baseline for the descender of g, on the baseline for H and x.

cc @pkgdemon