Normalize CRLF before highlighting source code

[sjh9714]

Summary

Refs #1037.

CRLF still appears as ^M inside highlighted comments and multiline strings, including in show_source and the source shown around a binding. Normalize CRLF before parsing colored source so those line endings render normally. Standalone carriage returns remain visible, and uncolored output is unchanged.

Changes

• Normalize CRLF in the shared source highlighter.
• Cover CRLF comments, multiline strings and method definitions, plus standalone CR handling.

Testing

• bundle exec rake test TEST=test/irb/test_color.rb
• bundle exec rake test (410 tests, 0 failures/errors, 2 omissions)
• bundle exec rubocop
• bundle exec rake build

Verified on macOS with Ruby 4.0.6. The Windows/Linux and other Ruby version matrices were not run locally.

[tompng]

Normalizing CRLF should be done AFTER Prism.parse_lex, not before. The original source code should be passed to the parser.
Replacing \r\n to \n will change the meaning of the code in some case, such as:

eval "%\r\r\na\r"
#=> "\na"
eval "%\r\r\na\r".gsub("\r\n", "\n")
#=> SyntaxError

eval "%\ra\r\r\n"
#=> "a"
eval "%\ra\r\r\n".gsub("\r\n", "\n")
#=> SyntaxError

Reline::Unicode.escape_for_print(code) is the one changing "\r" to "^M".
We can replace Reline::Unicode.escape_for_print(s) to a method call, calling a method defined in IRB::Color that escapes control characters except "\r\n".

[sjh9714]

Moved the normalization after parsing, as you suggested, and added both examples to the tests. I also fixed the failure with the latest Prism, which splits CRLF at the end of a comment across two tokens. The full suite passes locally with both released and development Prism.