feat(track-core): add portable training runtimes

Move scheme and PID training runtime behavior into the pure track_core layer and expose render sinks for injected strip application.

Add ESP compatibility adapters, Python bindings/test scaffolding, and in-memory render support so app_track_bt can consume core render and runtime logic without duplicating it.

Cover circular/linear rendering boundaries, all scheme runtime types, scheme render_to parity, PID sample de-duplication, speed suppression, and live tuning in track-core tests.

include/track_core/render.hpp

@@ -1,6 +1,7 @@
 #pragma once
 
 #include <array>
+#include <cstddef>
 #include <cstdint>
 
 #include "track_core/model.hpp"
@@ -27,10 +28,34 @@ struct TrackRenderPlan {
     std::size_t span_count{};
 };
 
+struct TrackRenderSink {
+    using ClearFn = TrackError (*)(void *context);
+    using FillFn = TrackError (*)(
+        void *context,
+        std::uint16_t start_led,
+        std::uint16_t led_count,
+        Color color);
+    using ShowFn = TrackError (*)(void *context);
+
+    void *context{};
+    ClearFn clear{};
+    FillFn fill{};
+    ShowFn show{};
+};
+
 [[nodiscard]]
 TrackRenderPlan make_track_render_plan(
     const TrackConfig &config,
     const TrackInfo &info,
     const TrackReport &report);
 
+[[nodiscard]]
+TrackError clear_render_sink(TrackRenderSink sink);
+
+[[nodiscard]]
+TrackError apply_render_plan(TrackRenderSink sink, const TrackRenderPlan &plan);
+
+[[nodiscard]]
+TrackError show_render_sink(TrackRenderSink sink);
+
 } // namespace track_core