bug#52670: legacy base64 encoding of latin-1

[mattiase@acm.org]

For what appears to be historical reasons, the base64 encoding functions (base64-encode-string etc) treat characters in the range U+0080..U+00FF as if they were raw bytes in the 127..255 range. This means that

  (base64-encode-string "ÿ")

and

  (base64-encode-string "\xff")

return the same result although the strings are completely different. Attempts to encode other multibyte characters fail (correctly). For example,

  (base64-encode-string "Ÿ")

signals an error, as expected.

I propose we tighten up the behavior by eliminating the legacy handling of characters in the  U+0080..U+00FF range. Letting the bug stay in place enables incorrect, brittle and error-prone usage: the functions are clearly intended to be fed encoded text only and should signal an error when not, as stated in the manual.