Fix COLR glyph extents calculations

[valadaptive]

I added custom tests for COLR glyph extents using the font mentioned in the TODO (see #157). Many of the tests failed, so I fixed the bugs causing them to do so:

• The union_ and intersect methods were copied incorrectly from HarfBuzz and overwrote the destination bounds entirely.
• We previously bailed out early if the COLR table was version 0, but we don't actually need to do this.
• COLRv0 glyphs in general (which can seemingly appear even if the version of the table is 1?) currently don't work because of an issue in ttf-parser (Call push_clip() and pop_clip() when painting COLRv0 glyphs ttf-parser#185). I've included tests for them anyway, which can be enabled if/when that PR is merged. I've confirmed that they pass when using a local ttf-parser with that change included.
• We previously bailed out entirely if we couldn't get the extents for a COLR glyph from the COLR table, but HarfBuzz falls back to looking at other tables.
• HarfBuzz intersects clip bounds with existing ones when push_clip is called. This doesn't seem to affect any tests, but I ran into it while tracking down the harder-to-spot first issue (union_ and intersect).

[LaurenzV]

That seems good, thanks!