fix(parser): handle heredoc pipe ordering and edge cases

[chaliy]

Summary

• Fix heredoc pipe ordering: cat <<EOF | sort now correctly pipes heredoc content to the next command in the pipeline
• Fix pipe continuation: cat <<EOF |\nsort works via rest-of-line re-injection
• Handle partial quote delimiters: <<'EOF'"2" correctly combines to delimiter EOF2 with quoting preventing expansion
• Fix multiple heredocs on one line: while cat <<E1 && cat <<E2; do ... done parses and executes correctly
• Fix heredoc followed by multiline dquote: cat <<EOF; echo "two\nthree" handles quoted strings spanning lines
• Capture while/until condition stdout so heredocs in loop conditions produce visible output

All 6 previously-skipped heredoc-edge spec tests are now enabled and passing.

Approach

Lexer: Added a VecDeque<char> re-injection buffer. read_heredoc saves the rest of the command line (after the heredoc delimiter token) instead of discarding it, then re-injects it after reading the heredoc body. peek_char/advance check the buffer first.

Lexer: read_continuation_into concatenates adjacent quoted/unquoted segments after a single-quoted string, enabling partial-quote heredoc delimiters.

Interpreter: execute_while and execute_until now capture and emit condition command stdout/stderr.

Test plan

• All 6 heredoc-edge skipped tests enabled and passing
• 2 new parser unit tests (heredoc pipe, multiple heredocs)
• Full spec test suite passes (1305 tests, 1192 passed, 0 failed, 113 skipped)
• bash_comparison_tests pass (identical output to real bash)
• All 1017 lib unit tests pass
• cargo fmt --check clean
• No new clippy warnings (pre-existing resolve_redirect_url dead_code only)

Closes #359