fix: stable let in

- I was using the previous position to compute the new
  position, but this causes the let in to not converge
  into a single round
- This way let in converge into a single round, as promised
  in the README

src/children.rs

@@ -20,9 +20,12 @@ impl Children {
 
         for child in node.children_with_tokens() {
             match child {
-                rnix::SyntaxElement::Node(_) => {
-                    children
-                        .push(Child { element: child, pos: pos.clone() });
+                rnix::SyntaxElement::Node(node) => {
+                    children.push(Child {
+                        element: node.clone().into(),
+                        pos:     pos.clone(),
+                    });
+                    pos.update(&node.text().to_string());
                 }
                 rnix::SyntaxElement::Token(token) => {
                     match token.kind() {

src/rules/let_in.rs

@@ -82,7 +82,7 @@ pub fn rule(
 
     // in
     let child = children.get_next().unwrap();
-    let indent = child.pos.column > 1;
+    let indent = build_ctx.pos_new.column > 1;
     match layout {
         crate::config::Layout::Tall => {
             steps.push_back(crate::builder::Step::Dedent);