Prior to this patch, the normalization functions in the
unicode module would raise a function clause error for
non-utf8 binaries.

This patch changes it so it returns {error, SoFar, Invalid}
as characters_to_binary and characters_to_list does in
the unicode module.

Note string:next_codepoint/1 and string:next_grapheme had
to be changed accordingly and also return an error tuple.

A base for unicode functions, not intended to be a user api.

Whitespace returns a reasonable subset of non nobreak whitespace
characters.

Implementation notes:

Make function clauses instead of using arrays and store tuples instead
of maps to save space.