GitLab

In some languages, the current logic (iteration over code
points and then get the glyphs by CharRangeToGlyphRange)
makes wrong effect, because both a character and its
diacritic marks can point to the same glyph range and
it misunderstands the width of the word for wrapping.

This CL changes the iteration logic for clusters so that
it skips the code points for the diacritic marks.

BUG=470073
R=msw@chromium.org, ckocagil@chromium.org
TEST=the new test case covers

Review URL: https://codereview.chromium.org/1036663003

Cr-Commit-Position: refs/heads/master@{#322316}