fix(bash): keep heredoc continuations highlighted

[puneetdixit200]

Fixes #4377.

Summary

• Keep Bash heredoc continuation commands on the command line instead of folding them into the heredoc string.
• Start the heredoc string span on the following line and end it at the matching delimiter.
• Add a Bash markup regression for <<-EOF | cat.
• Update the existing Bash heredoc expectation and changelog entry.

Tests

• Red before fix: the reported sample highlighted EOF | cat inside the heredoc string span, so cat was not highlighted as a command.
• ONLY_LANGUAGES=bash npm run test-markup
• npx eslint --no-eslintrc -c .eslintrc.lang.js src/languages/bash.js
• npm run build
• git diff --check

Note: I also ran the full npm run test-markup; it reached 537 passing tests and then hit the existing 2s Mocha timeout on http/default, unrelated to the Bash fixture. The Bash markup suite passed in both focused runs.

[joshgoebel]

All of this is because of the pipe and piping the string into any sequence of bash, right?

[arturict]

Not only pipes. The same split is needed for any shell syntax after the delimiter word, for example <<EOF >out or <<EOF && echo queued. The body should start at the following newline; everything before that newline should remain normal Bash.

[joshgoebel]

What is the requirement for heredoc naming in Bash? Could this be tightened up any further than \w+?

[arturict]

I think this needs broadening rather than tightening. Bash takes a word after << and applies quote removal; it is not limited to \w+.

For example, this branch captures only END here and never finds the real terminator, so echo done stays inside the string:

cat <<END-OF | cat
hello
END-OF
echo done

<<'EOF' is valid too, but is not recognized as a heredoc by this matcher. Could we add both as markup cases before settling on the delimiter match?

[joshgoebel]

Sure. Add all the cases we know of, then try to get the highlighted correctly - that's the way to do it.

[arturict]

Nice rust escaping coverage. Could you include minimal unit case for escaped quotes/backslashes in char literals to aid review traceability?