Currently, in order to be consistent with Google SQL, the first-party implementations normalize unescaped CR (carriage return) and CRLF sequences to LF (line feed) in triple-quoted string literals.
My preference is to add conformance tests that fail on this (currently unspecified and untested) behavior, but alternatively, we'll need to update the spec to include it.
Per discussion with @TristonianJones:
My GoogleSQL translation checks seem to not result in any issues, but I'll need to do some broader tests to validate that I can drop the normalization in a timely fashion.
I'm opening this issue to track follow-up.