normalize literal \r and \r\n chars in string literals to \n

[stevengj]

This fixes #11988, by normalizing literal \r and \r\n characters in multiline string literals to \n. Explicitly escaped \r chars are unaffected.

Also fixes #10833, I think (unix2dos test/markdown.jl && make -C test markdown passes).

[stevengj]

@tkelman, is this something that should be backported? I would tend to view it as a bugfix; I'd be very suspicious of any code that relied on the old behavior.

[StefanKarpinski]

I agree that this should be backported.

[StefanKarpinski]

CI failure looks unrelated and incidental.

[tkelman]

This should probably have been left open for review a little longer than 2 hours. I'm not entirely convinced forcing this behavior at the parser level is right. Seems like the output bytes should maybe match the input bytes here, and string handling routines in Markdown and elsewhere should be made smarter wrt CRLF handling.

[tkelman]

This is probably fine, especially considering Python behaves this way, but it should really be documented as the way the language parses multiline strings - somewhat related to the handling of escape codes.

It also doesn't fix any issues stemming from Markdown or other string handling methods when reading from an external file, just hides the underlying issue for julia source files.

[stevengj]

I thought that there was a clear consensus in #11988 that this was the desired behavior, sorry I jumped the gun on the merge button. I'll push a documentation update.

Also, the hornéd god hath spoken upon this issue.