Lexer accidentally(?) does not use is_ascii_whitespace for literal whitespace in string continuations

[hkBst]

#108403 proposed to fix this, but it was claimed that the current behavior was documented in the reference in this comment. Incorrectly, as far as I can see, as that page only describes whitespace escapes as being \r, \t, and \n and the fix was about literal whitespace in string continuations. Now https://doc.rust-lang.org/reference/expressions/literal-expr.html#string-continuation-escapes does describe this behavior, but this was added later in Jan 2024. Indeed, this PR shows the reference documented skipping all whitespace, until Jun 13, 2022.

Current behavior has this ui test. It seems like this behavior was once implemented like it is now, then got claimed to be canon then got documented as canon. Anyway, I'm not sure why not all unicode whitespace is skipped, but just almost all ascii whitespace, but it seems important to pick an existing whitespace set, instead of using an old bad manual implementation of is_ascii_whitespace...

Perhaps we can see a crater run at least...

[jieyouxu]

cc @petrochenkov as I think you're familiar with this specific quirk

[petrochenkov]

I'm aware of it, but don't have any opinion on whether the behavior here should be changed or not (I also didn't do the archeology to find out why it is like it is).