feat(format): skip non-plpgsql bodies; keep literals and embedded code safe

- Only restyle bodies for plpgsql/sql; skip pl* (plpython3u, plperl, ...), c and internal
- Keep LANGUAGE clause on DO blocks
- Carry multi-line string and dollar-quoted literals verbatim in bodies and DML
- Format dollar-quoted literals that contain code (declare/begin/select/...), never format() templates
- Fix code joined onto -- comments, early flush after raise exception, dropped comment before BEGIN
- Regenerate test_mm_proc golden; document in README, plan, todo

pkg/format/format.go

@@ -55,7 +55,7 @@ func (p *printer) writeItem(n cst.Node) {
 	case *cst.Raw:
 		switch {
 		case isDMLStart(v.Toks):
-			p.b.WriteString(formatDML(v.Toks, p.st))
+			p.b.WriteString(restoreLiteralNewlines(formatDML(v.Toks, p.st)))
 		case isDoBlock(v.Toks):
 			p.b.WriteString(formatDoBlock(v.Toks, p.st))
 		default:
@@ -82,8 +82,10 @@ func isDoBlock(toks []cst.Tok) bool {
 // formatDoBlock formats a DO $$ ... $$ block by applying formatBody to the
 // dollar-quoted string and emitting DO + newline + formatted body.
 func formatDoBlock(toks []cst.Tok, st config.Style) string {
-	// Find the DO keyword, the dollar-string body, and the optional semicolon.
+	// Find the DO keyword, the dollar-string body, any other clause tokens
+	// (LANGUAGE x, before or after the body), and the optional semicolon.
 	var doTok, bodyTok *cst.Tok
+	var pre, post []string
 	hasSemi := false
 	for i := range toks {
 		t := &toks[i]
@@ -101,6 +103,21 @@ func formatDoBlock(toks []cst.Tok, st config.Style) string {
 		}
 		if t.Tok.Kind == lexer.Semicolon {
 			hasSemi = true
+			continue
+		}
+		// Clause tokens other than DO / body / ';' (e.g. LANGUAGE plpython3u).
+		// Any comment attached to one means we cannot safely relocate it.
+		if len(t.Comments()) > 0 {
+			return verbatimSpanFormatBody(toks, bodyTok, st)
+		}
+		txt := t.Tok.Text
+		if low == "language" {
+			txt = applyCase(txt, st.KeywordCase)
+		}
+		if bodyTok == nil {
+			pre = append(pre, txt)
+		} else {
+			post = append(post, txt)
 		}
 	}
 	if doTok == nil || bodyTok == nil {
@@ -110,8 +127,20 @@ func formatDoBlock(toks []cst.Tok, st config.Style) string {
 	nl := st.Newline
 	var b strings.Builder
 	b.WriteString(applyCase(doTok.Tok.Text, st.KeywordCase))
+	if len(pre) > 0 {
+		b.WriteString(" ")
+		b.WriteString(strings.Join(pre, " "))
+	}
 	b.WriteString(nl)
-	b.WriteString(formatBody(bodyTok.Tok.Text, st))
+	if isPlpgsql(toks) {
+		b.WriteString(formatBody(bodyTok.Tok.Text, st))
+	} else {
+		b.WriteString(bodyTok.Tok.Text)
+	}
+	if len(post) > 0 {
+		b.WriteString(nl)
+		b.WriteString(strings.Join(post, " "))
+	}
 	if hasSemi {
 		b.WriteString(";")
 	}
@@ -193,7 +222,11 @@ func (p *printer) writeCreateFunction(cf *cst.CreateFunction) {
 	}
 	if cf.Body != nil {
 		p.nl()
-		p.b.WriteString(formatBody(cf.Body.Tok.Text, p.st))
+		if isPlpgsql(cst.Tokens(cf)) {
+			p.b.WriteString(formatBody(cf.Body.Tok.Text, p.st))
+		} else {
+			p.b.WriteString(cf.Body.Tok.Text)
+		}
 	}
 	for _, clause := range cf.Tail {
 		p.nl()
@@ -445,7 +478,7 @@ func verbatimSpanFormatBody(toks []cst.Tok, bodyTok *cst.Tok, st config.Style) s
 				b.WriteString(tr.Text)
 			}
 		}
-		if bodyTok != nil && t.Tok.Kind == lexer.DollarString && t.Tok.Off == bodyTok.Tok.Off {
+		if bodyTok != nil && t.Tok.Kind == lexer.DollarString && t.Tok.Off == bodyTok.Tok.Off && isPlpgsql(toks) {
 			b.WriteString(formatBody(t.Tok.Text, st))
 		} else {
 			b.WriteString(t.Tok.Text)
@@ -546,3 +579,27 @@ func alignParamTypes(params []string) []string {
 	}
 	return out
 }
+
+// isPlpgsql reports whether the routine's body should be restyled. Bodies in
+// any pl* language other than plpgsql (plpython3u, plperl, pltcl, ...), and in
+// c or internal, are emitted verbatim; plpgsql, sql and everything else are
+// formatted. A routine
+// with no LANGUAGE clause is treated as sql and formatted.
+func isPlpgsql(toks []cst.Tok) bool {
+	var sig []cst.Tok
+	for _, t := range toks {
+		if !t.Tok.IsTrivia() && t.Tok.Kind != lexer.EOF {
+			sig = append(sig, t)
+		}
+	}
+	for i := 0; i+1 < len(sig); i++ {
+		if sig[i].Is("language") {
+			name := strings.ToLower(strings.Trim(sig[i+1].Tok.Text, "'\""))
+			if name == "c" || name == "internal" {
+				return false
+			}
+			return !strings.HasPrefix(name, "pl") || name == "plpgsql"
+		}
+	}
+	return true
+}