优化原生结果编码传输与浏览器缓存，记录八路性能基线

原生结果series通过字节索引直传，C端使用Ryu精确回读编码和64 KiB批量写出；网页采用Float64缓存和CSV工作线程，减少结果处理与保存等待。

补充八路AME曲线核查、全流程分阶段计时、独立编码基准和复现工具，固定后续优化采用修正八路及rtol=1e-8。C写出1.1808→0.1638 s，点击到可查看8.0100→6.9756 s。

验证：最终10项编码专项、29项相关后端回归通过；8份原生结果逐位一致，16次网页结果/CSV/刷新恢复通过。前端构建及缓存/CSV专项在本轮结果处理工作中通过。环境、原始大结果与临时构建不纳入Git。

native/include/json_numbers.h

@@ -0,0 +1,20 @@
+#ifndef NATIVE_JSON_NUMBERS_H
+#define NATIVE_JSON_NUMBERS_H
+#include <stddef.h>
+#include <stdio.h>
+
+#define NATIVE_JSON_DOUBLE_CAPACITY 32
+
+/* Locale-independent finite binary64 token; no terminating NUL is promised.
+ * Returns its length, or zero for NaN/Infinity. Negative zero is -0.0 so both
+ * Python and JavaScript JSON readers preserve its sign. */
+int native_json_format_double(char output[NATIVE_JSON_DOUBLE_CAPACITY], double value);
+
+/* Write a number, or a complete array from values[i*stride]. Each call flushes
+ * its own fixed-size buffer into FILE, but does not fflush/fclose the FILE.
+ * Return zero on invalid input/nonfinite values/write failure. A failed array
+ * may have written a prefix: callers must not publish that document as valid.
+ * The caller owns FILE and must check its final flush/close before publishing. */
+int native_json_write_number(FILE *file, double value);
+int native_json_write_array(FILE *file, const double *values, size_t count, size_t stride);
+#endif