[ttx] Non-ascii in CFF FullName string doesn't round-trip

A font mentioned in #2897 uses non-ascii in the CFF FullName field. This is somehow dumped to .ttx with an unclear encoding, and reading the dumped .ttx fails on the non-ascii chars:

Compiling "/Users/just/Downloads/LeeSeoyun.ttx" to "/Users/just/Downloads/LeeSeoyun#1.otf"...
Parsing 'GlyphOrder' table...
Parsing 'head' table...
Parsing 'hhea' table...
Parsing 'maxp' table...
Parsing 'OS/2' table...
Parsing 'name' table...
Parsing 'cmap' table...
Parsing 'post' table...
Parsing 'CFF ' table...
ERROR: Unhandled exception has occurred
Traceback (most recent call last):
  File "/Users/just/code/git/fonttools/Lib/fontTools/ttx.py", line 405, in main
    process(jobs, options)
  File "/Users/just/code/git/fonttools/Lib/fontTools/ttx.py", line 387, in process
    action(input, output, options)
  File "/Users/just/code/git/fonttools/Lib/fontTools/misc/loggingTools.py", line 372, in wrapper
    return func(*args, **kwds)
  File "/Users/just/code/git/fonttools/Lib/fontTools/ttx.py", line 298, in ttCompile
    ttf.importXML(input)
  File "/Users/just/code/git/fonttools/Lib/fontTools/ttLib/ttFont.py", line 349, in importXML
    reader.read()
  File "/Users/just/code/git/fonttools/Lib/fontTools/misc/xmlReader.py", line 47, in read
    self._parseFile(self.file)
  File "/Users/just/code/git/fonttools/Lib/fontTools/misc/xmlReader.py", line 72, in _parseFile
    parser.Parse(chunk, 0)
  File "/Users/sysadmin/build/v3.10.5/Modules/pyexpat.c", line 470, in EndElement
  File "/Users/just/code/git/fonttools/Lib/fontTools/misc/xmlReader.py", line 155, in _endElementHandler
    self.currentTable.fromXML(name, attrs, content, self.ttFont)
  File "/Users/just/code/git/fonttools/Lib/fontTools/ttLib/tables/C_F_F_.py", line 46, in fromXML
    self.cff.fromXML(name, attrs, content, otFont)
  File "/Users/just/code/git/fonttools/Lib/fontTools/cffLib/__init__.py", line 346, in fromXML
    topDict.fromXML(name, attrs, content)
  File "/Users/just/code/git/fonttools/Lib/fontTools/cffLib/__init__.py", line 2618, in fromXML
    value = conv.xmlRead(name, attrs, content, self)
  File "/Users/just/code/git/fonttools/Lib/fontTools/cffLib/__init__.py", line 1358, in xmlRead
    return tobytes(attrs["value"], encoding=("ascii"))
  File "/Users/just/code/git/fonttools/Lib/fontTools/misc/textTools.py", line 131, in tobytes
    return s.encode(encoding, errors)
UnicodeEncodeError: 'ascii' codec can't encode characters in position 0-11: ordinal not in range(128)

OTMaster somehow manages to read this field as a Korean string (which makes sense, given the font):

Yet fonttools reads it as gibberish: 'ì\x9d´ì\x84\x9cì\x9c¤ì²´'.

[schriftgestalt]

The current version should make sure that it contains ASCII only.

[justvanrossum]

but encoding a previously utf-8 string to latin1 could fail, or decoding a utf-8 string as latin-1 give you gibberish - how would that round-trip?

We start with a byte string without knowing the encoding. Converting this to Latin-1 can’t fail. Encoding the converted Latin-1 string as utf-8 can’t fail either. This works backwards as well. Yes, the XML will contain gibberish, but it does round-trip:

s = "".join(chr(i) for i in range(256))
b = s.encode("latin-1")
assert list(b) == list(range(256))
utf = s.encode("utf-8")
assert utf.decode("utf-8") == s

b2 = bytes(range(256))
s2 = b2.decode("latin-1")
assert s2 == s

Not saying this is a nice solution, but given there is no encoding information at all, and given any non-ascii in FullName can be considered broken, this at least would round-trip without error.

And I agree that if we can plug in that write8bit functionality here, it would be the better and cleaner solution.