Text layout should refer to character indices, not byte offsets

[burg]

In general, all text layout should be in terms of character offsets, and refer to/save data in glyph storage for determining ligatures, clusters, and line breaking. Line breaking is currently done on whitespaces in the original string (~[u8]), which causes inter-word breaks when there is not a 1-to-1 correspondence between bytes, glyphs, and/or characters.

This should fix bad line breaking, invalid indexing into the glyph store, and unblock rendering of multi-byte characters. A Japanese test case is src/tests/mojira.html.

[mbrubeck]

This has been fixed for a while now.