Optimize glyph row clearing and copying routines

[Eli Zaretskii]

Does this change really speed up the code?  AFAIU, previously the
struct assignment could use word-size copies (because a struct is
always aligned), but now you cast the arguments to 'char *' and use
memcpy, which could fall back on copying single bytes or shorter
words.

Did you measure the impact?  Was it significant enough to justify this
change?