The width argument of `T2CharStringPen()` is inserted directly into the
CharString program, so it must be relative to Private.nominalWidthX, but
CharString.width is a calculated absolute value.
Some implementations, notably Adobe’s, will use the width from the CFF
CharString instead of the one from hmtx table.
Fixes https://github.com/fonttools/fonttools/issues/3658
This is to be able to ignore tricky glyphs that sometimes trip up Skia PathOps.Simplify operation.
We have no idea how to fix this upstream (short of having the glyphs redrawn), and perfect is the enemy of good..
Sometimes skia-pathops simplify may fail (for unknown reasons which I'm still trying to debug).
It's a good idea to know the name of the offending glyph
https://github.com/google/fonts/issues/3365
the order of contours modified by Skia is not stable and may change. So when we compare the original and the modified paths, we compare the paths as unordered sets of contours.
The order of contours doesn't produce any visible difference, but we try to keep the changes to the minimum here to avoid unnecessary diffs
