Avoid per-label allocations and redundant stable sorting

src/core/label_fade_tracker.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include <algorithm>
+#include <array>
 #include <chrono>
 #include <string>
 #include <type_traits>
@@ -146,11 +147,25 @@ bool update_and_draw_faded_labels(
         bool               primary    = false;
         bool               suppressed = false;
     };
+    class Emitted_texts
+    {
+    public:
+        void push_back(emitted_text_t text) { m_values[m_count++] = text; }
+        auto begin()                       { return m_values.begin(); }
+        auto end()                         { return m_values.begin() + m_count; }
+        auto begin() const                 { return m_values.begin(); }
+        auto end() const                   { return m_values.begin() + m_count; }
+
+    private:
+        // A label only emits its previous and current text during a crossfade.
+        std::array<emitted_text_t, 2> m_values{};
+        std::size_t                  m_count = 0;
+    };
     struct emitted_state_t
     {
         key_t                       key;
         const std::string*          final_text = nullptr;
-        std::vector<emitted_text_t> texts;
+        Emitted_texts               texts;
     };
 
     bool                         any_active = false;

src/core/series_renderer.cpp

@@ -1524,7 +1524,7 @@ void Series_renderer::prepare(
                     layer->z_order()});
             }
         }
-        // Sort by z_order before preparing. The global stable_sort of
+        // Sort by z_order before preparing. The global sort of
         // prepared_draws below independently fixes the final record order, but
         // this per-view sort is still load-bearing: it determines the order in
         // which layers are *prepared*, which is a tested contract (a lower
@@ -1786,7 +1786,8 @@ void Series_renderer::prepare(
         }
     }
 
-    std::stable_sort(
+    // The unique insertion_order breaks all ties, so stability adds no ordering guarantee.
+    std::sort(
         m_rhi_state->prepared_draws.begin(),
         m_rhi_state->prepared_draws.end(),
         [](const auto& a, const auto& b) {