With the turtle graphics, some very long lines might be returned by Poseidon just after the CR (\r). In this case, our new normalization will fix the mistake, but the next message received (just a LF `\n`) will be forwarded to the clients unchanged. This is not desired, so that we filter this case separately.
After splitting the messages received into lines, we could still have a line break within quotes. Since we just want to modify terminating line breaks, we change the recognition accordingly.
In Python, a so-called doc string consists of a string encoded in three quote (starting and ending with """ or '''). These strings were not yet recognized correctly, so that line breaks there would cause further trouble with recognizing individual lines (which is important for the linter translation).
Previously, a message received with CRLF line endings was (incorrectly) converted to CRCRLF line endings (i.e., from \r\n to \r\r\n). Then, the splitting in individual lines could generate an "empty" line, just containing the newly-added CR line ending \r.
If a non-JSON message received contains quotes and is cut off weirdly, the following message might "quote" the line breaks. Then, we have an issue and cause \r\n\n for the further processing within the CodeOcean backend.
This change is required for Turtle output generated by the Python environment. Otherwise, drawing some images could add empty new lines (in the browser), since the original JSON message got split between the ending \r\n characters.
* This is based on Sentry issue CODEOCEAN-5W (linter translation failing due to \r in output)
* We cannot keep all "\r" messages but we must not forward the "\r" if it is a single message received after a valid JSON
