Forbid base64(string) in favor of base64(array)

[sbarzowski]

Applying base64 on string uses a weird encoding. I think we should fail with an error explaining what to use instead (and how to get the old behavior by naively encoding the string).

Right now it's a major footgun, so I think breaking backwards compatibility is worth it here (especially since the fix is easy on the user side).

[sparkprime]

To be clear, the current behavior encodes it as Latin1, is that right? And if it's not within Latin1 the unicode will generate higher codepoints and it will error. Whereas a more sensible default is to encode it with UTF-8, which would yield the same if it's ASCII but different within Latin 1 (and actually produce something useful outside of Latin1)

[sbarzowski]

Currently base64 takes the unicode codepoints. If it's greater than 255 it errors out. It seems like Unicode codepoints <=255 are the same as the original latin-1 (aka ISO 8859-1), yes. I didn't notice that before. So at least it's a fairly well-known encoding.

Yes, a more sensible default is to encode it with UTF-8. Our strings contain Unicode after all. We don't really have that option – that could silently break users. My proposition is to always require an explicit encoding (e.g. using std.encodeUTF8).