修复循环信号与事件采样并接入 LSTP 接触定位，补充八路验证及复用实验

相较上一版 Jacobian 确定性复用更新，本次补齐事件边界一致性、结果两侧采样及接触事件定位；保留已有物性复用和组件力学公式。

- 统一 UD00 信号求值与下一事件查询的绝对时间边界，修复循环边界浮点舍入导致的阶段错位、重复或漏报，并覆盖零时长、多阶段及长周期场景。
- 引入原生输出语义 v2：保留规则网格真实时间，补充内部时间事件和状态事件的左邻及事件后采样，按保存时间、状态和离散模式重放结果。
- 两条代码生成路径均发出 LSTP 接触描述，默认定位间隙过零及非负力模式的力截断；仅在接受事件时更新防重复记录，增加 contactEvents 诊断计数。
- 补充 MASS/LSTP 独立事件实验、八路全曲线与驱动阶段配对评估，以及 Amesim 不连续点输出对照和力差定位报告；MASS 新增释放机制仍保留为独立实验。
- 保存局部 probe、context 访问与回退、shadow replay、R288 real skip/typed replay 及阀门数值尾部诊断工具和报告；未证明净收益的实验不启用为生产默认优化。
- 更新原生运行说明和元件建模规范，补充信号边界、输出语义、接触事件和实验依赖回归测试。

验证：五组专项回归共 34 项全部通过；37 个待提交 Python 文件语法检查通过；git diff --cached --check 通过。

tests/manual/local_probe_profile.h

@@ -0,0 +1,53 @@
+/* Diagnostic-only exclusive buckets; included only by the copied worker. */
+#ifndef LOCAL_PROBE_PROFILE_H
+#define LOCAL_PROBE_PROFILE_H
+#ifndef _WIN32_WINNT
+#define _WIN32_WINNT 0x0600
+#endif
+#include <windows.h>
+#include <stdint.h>
+#include <cpuid.h>
+#include "local_probe.h"
+enum { PF_OTHER, PF_INCREMENT, PF_PERTURB, PF_ZERO, PF_ASSEMBLY,
+       PF_BASELINE, PF_INIT, PF_GAS_PREP, PF_RETAINED, PF_FALLBACK,
+       PF_CONTEXT, PF_SNAPSHOT, PF_RESTORE, PF_NODE, PF_PORT,
+       PF_MECHANICAL, PF_GAS_EQUATIONS, PF_OUTPUTS, PF_PIPE,
+       PF_FINITE, PF_DISPATCH, PF_FULL_FALLBACK, PF_CATEGORIES };
+/* row 0 = outer callback, row 1 = baseline, row 2+g = probe group g. */
+typedef struct { uint64_t ticks,intervals; } PfBucket;
+extern PfBucket pf_buckets[LP_NC+2][PF_CATEGORIES],*pf_current;
+extern uint64_t pf_previous_tick;
+extern unsigned pf_row;
+extern int pf_coarse;
+extern const unsigned char pf_affected[LP_NC][LP_NO];
+extern uint64_t pf_operations[LP_NC][2];
+static inline uint64_t pf_clock(void){
+    unsigned lo,hi;
+    __asm__ __volatile__("lfence\n\trdtsc\n\tlfence" : "=a"(lo),"=d"(hi) :: "memory");
+    return ((uint64_t)hi<<32)|lo;
+}
+static inline void pf_mark(unsigned category) {
+    uint64_t t=pf_clock();
+    pf_current->ticks+=t-pf_previous_tick;
+    pf_current->intervals++;
+    pf_previous_tick=t;pf_current=&pf_buckets[pf_row][category];
+}
+#define PF_MARK(category) pf_mark(category)
+#define PF_SCOPE(row,category) do {pf_row=(row);pf_mark(category);} while(0)
+#define PF_PROBE(category) do {if(lp_color>=0)pf_mark(category);} while(0)
+void pf_initialize(void);void pf_finish(void);
+int pf_select(void);void pf_flush(uint64_t,uint64_t);
+void pf_eval_count(void);
+static inline void pf_operation(int position){
+    int retained=pf_affected[lp_color][position];
+    pf_operations[lp_color][retained?0:1]++;
+    PF_MARK(retained?PF_RETAINED:PF_FALLBACK);
+}
+void pf_open(void);void pf_close(void);
+void pf_snapshot(int,NativePropertyCache*,NativePipeCache*);
+void pf_save(double*,double*,double*,double*,double*);
+int pf_reuse(int,NativePropertyCache*,NativePipeCache*,double*,double*,double*,double*,double*);
+int pf_lp_eval(double,const double*,double*,double*,ModelJacobianWorkspace*);
+void lp_jac_time_end(uint64_t,uint64_t);
+void pf_validate_matrix(double,const double*,const double*);
+#endif