fix(format): align cmdsub reformatter with bash canonical form

[mpecan]

Summary

Fixes the six canonical-reformatter drifts in $(…) body formatting tracked by #40.

bug	fix
{fd}> dropped	new Redirect.varfd: Option<String> AST field + parse_redirect_with_varfd + formatter emits {name} prefix
select rendered as raw S-expr	new format_select arm, shares layout helper with format_for
function body not wrapped in implicit braces	format_function_body rewritten to always emit { … }
statements in function body single-lined	new format_list_block helper — wraps on ;/& but keeps &&/`
<<\EOF quoted-delimiter lost	honor HereDoc.quoted in both regular and heredoc-pipe paths

build_redirect hit the 60-line limit with the new field, so its heredoc branch was factored into build_heredoc_redirect. format_for and format_select delegate to a shared format_for_or_select helper to remove duplication.

Parable corpus correction

Three tests in tests/parable/12_command_substitution.tests expected the reformatter to strip heredoc delimiter quoting (<<'EOF' → <<EOF). Parable was wrong about bash's canonical form — verified directly against bash-oracle --dump-ast:

$ printf "x=\$(cat <<'EOF'\nbody\nEOF\n)\n" | bash-oracle --dump-ast
(command (word "x=$(cat <<'EOF'\nbody\nEOF\n)"))

Updated the 3 goldens to match the correct quoted form.

Bonus fix

The heredoc-quoting fix also newly passes heredoc_in_cmdsub 1 (tracked under #39), removed from KNOWN_ORACLE_FAILURES with an attribution comment.

Test plan

• cargo fmt
• cargo clippy --all-targets -- -D warnings — no warnings
• cargo test — 240 passed, 0 failed (185 lib + 54 integration + 1 tree-sitter compare)
• All 6 cmdsub_reformat N cases now pass in oracle_bash_valid_divergences
• heredoc_in_cmdsub 1 also newly passes
• New parser unit tests pin Redirect.varfd populated vs. None
• 3 new corpus regressions in tests/parable/12_command_substitution.tests: tab-stripped quoted heredoc, && in function body, varfd close-fd
• Full test suite green

Closes #40.