layout: restructure into src/ tests/ android/ scripts/ tools/

- extension/src/{port,platform,codepage} -> src/; native_render -> src/host
  (+ dxt, shaders/); libgr2 -> src/gr2; extension/third_party -> third_party
- extension/tests -> tests/port, libgr2/tests -> tests/gr2
- android-native -> android (build.sh, push-client.sh moved in)
- script -> scripts; tools/40250 -> tools/server; oracle -> tools/granny-oracle;
  perf tools -> tools/perf
- all build trees under build/ (native, release, android, port-gate)
- xrender:: CMake aliases -> mt::; port-map ledger impl paths rewritten

No code changes. ctest 15/15, port_gate macos+android PASS, port_map check 0 errors.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/host/dxt.h

@@ -0,0 +1,27 @@
+// dxt.{h,cpp} — ported verbatim from xrender-poc/engine/dxt.{h,cpp}
+// (namespace engine -> mtimage). DDS DXT1/3/5 + masked 16/24/32-bit RGB -> RGBA8, level 0.
+// engine/dxt —— DDS(DXT1/3/5) → RGBA8 软解。见 docs/steps/M1-static-render.md T3
+// 自己写 ~150 行绕开 reuse/EterImageLib 的 windows.h 依赖（M1 风险表允许）。
+// 对拍时两侧都喂软解 RGBA，不被 GPU S3TC 的 bit 级差异污染。
+#pragma once
+#include <cstdint>
+#include <vector>
+
+namespace mtimage {
+
+struct Image {
+    uint16_t w = 0, h = 0;
+    std::vector<uint8_t> rgba;   // w*h*4，level 0
+    // DXT DDS only: level count 40250 CreateDDSTexture gives the texture (0 = not a DXT
+    // DDS, caller follows the D3DX_DEFAULT full-chain path) and levels 1.. decoded to RGBA8.
+    uint8_t file_mip_levels = 0;
+    std::vector<std::vector<uint8_t>> mips;
+    const char* format = "";     // "DXT1" / "DXT3" / "DXT5" / "BGRA8" / ""
+    bool ok() const { return w && h && rgba.size() == size_t(w) * h * 4; }
+};
+
+// 读整个 .dds 文件字节，解 level 0 到 RGBA8（DXT 另解文件内 mip 到 mips）。失败返回 !ok()。
+Image load_dds(const uint8_t* bytes, size_t len);
+Image load_dds_path(const char* path);
+
+} // namespace mtimage