Fix and clarify CR LF normalization and CR in string literals

[LukasKalbertodt]

This was slightly incorrect before. Relevant commits changing this:

• fa56fdb
• 27e1ec9

The normalization is not applied repeatedly, so CR LF pairs can still exist. Further, given that the normalization happens before lexing, the part "other than as part of such a string continuation escape" is not useful. Either it was CR LF in the raw input, but has already been transformed already (so the lexical grammar does not see CR). Or there is a surviving CR LF pair after the normalization, which is disallowed tho.

Here are two test programs showing this behavior:

printf 'fn main() { "a\r\r\n\nb"; }' > code.rs | rustc -

Results in:

error: bare CR not allowed in string, use `\r` instead
 --> <anon>:1:15
  |
1 | fn main() { "a␍
  |               ^
  |
help: escape the character
  |
1 | fn main() { "a\r
  |               ++

And

printf 'fn main() { "a\\\r\r\n\nb"; }' > code.rs | rustc -

Results in

error: unknown character escape: `\r`
 --> <anon>:1:16
  |
1 | fn main() { "a\␍
  |                ^ unknown character escape
  |
  = help: this is an isolated carriage return; consider checking your editor and version control settings

[ehuss]

Thanks!