Optimize FlatText.get/3 memory usage and speed (#664)

* Optimize FlatText.get/3 memory usage and speed

Replaced Enum.reduce/3 with an optimal tail-recursive approach in FlatText.

By traversing lists manually, we avoid the overhead of the Enumerable protocol and the allocation of anonymous closures per reduction iteration.

Benchmark results when running Floki.FlatText.get on a large document:

```text
Name                     ips        average  deviation         median         99th %
Tail recursion        3.21 M      311.16 ns ±23057.88%          50 ns         160 ns
Enum.reduce           2.76 M      361.71 ns ±20499.20%          50 ns         180 ns

Comparison:
Tail recursion        3.21 M
Enum.reduce           2.76 M - 1.16x slower +50.55 ns

Memory usage statistics:

Name              Memory usage
Tail recursion           136 B
Enum.reduce              208 B - 1.53x memory usage +72 B
```

* Add tests for FlatText edge cases

* Run formatter

Author: preciz

lib/floki/flat_text.ex

@@ -17,9 +17,7 @@ defmodule Floki.FlatText do
 
   def get(html_nodes, sep, include_inputs?) when is_list(html_nodes) do
     html_nodes
-    |> Enum.reduce([], fn html_node, acc ->
-      text_from_node(html_node, acc, 0, sep, include_inputs?)
-    end)
+    |> text_from_nodes([], 0, sep, include_inputs?)
     |> IO.iodata_to_binary()
   end
 
@@ -29,6 +27,13 @@ defmodule Floki.FlatText do
     |> IO.iodata_to_binary()
   end
 
+  defp text_from_nodes([], acc, _, _, _), do: acc
+
+  defp text_from_nodes([node | rest], acc, depth, sep, include_inputs?) do
+    acc = text_from_node(node, acc, depth, sep, include_inputs?)
+    text_from_nodes(rest, acc, depth, sep, include_inputs?)
+  end
+
   defp text_from_node({"input", attrs, []}, acc, _, _, true) do
     [acc, Floki.TextExtractor.extract_input_value(attrs)]
   end
@@ -39,9 +44,7 @@ defmodule Floki.FlatText do
 
   defp text_from_node({_tag, _attrs, html_nodes}, acc, depth, sep, include_inputs?)
        when depth < 1 do
-    Enum.reduce(html_nodes, acc, fn html_node, acc ->
-      text_from_node(html_node, acc, depth + 1, sep, include_inputs?)
-    end)
+    text_from_nodes(html_nodes, acc, depth + 1, sep, include_inputs?)
   end
 
   defp text_from_node(text, [], _, _sep, _) when is_binary(text), do: text

test/floki/flat_text_test.exs

@@ -84,4 +84,43 @@ defmodule Floki.FlatTextTest do
 
     assert Floki.FlatText.get(nodes, " ") == expected_text
   end
+
+  # --- Edge Cases ---
+
+  test "does not extract text from deeply nested nodes (depth >= 1 for text nodes)" do
+    node = {"div", [], [{"span", [], [{"p", [], ["Should not be here"]}]}]}
+    assert Floki.FlatText.get(node) == ""
+  end
+
+  test "does not extract inputs from deeply nested nodes (depth >= 1 for inputs)" do
+    node = {"div", [], [{"span", [], [{"input", [{"value", "hidden"}], []}]}]}
+    assert Floki.FlatText.get(node, " ", true) == ""
+  end
+
+  test "handles multiple consecutive text nodes at root level" do
+    nodes = ["one", "two", "three"]
+    assert Floki.FlatText.get(nodes) == "onetwothree"
+    assert Floki.FlatText.get(nodes, "-") == "one-two-three"
+  end
+
+  test "handles multiple consecutive text nodes inside a node" do
+    node = {"div", [], ["one", "two", "three"]}
+    assert Floki.FlatText.get(node) == "onetwothree"
+    assert Floki.FlatText.get(node, "-") == "one-two-three"
+  end
+
+  test "ignores comments and doctypes" do
+    nodes = [
+      {:comment, "this is a comment"},
+      "text",
+      {:doctype, "html", "", ""}
+    ]
+
+    assert Floki.FlatText.get(nodes) == "text"
+  end
+
+  test "empty nodes return blank string" do
+    assert Floki.FlatText.get([]) == ""
+    assert Floki.FlatText.get({"div", [], []}) == ""
+  end
 end