refactor(video-pipeline): 移除 segments 机制，改为整段音频合成

移除 TTS 阶段逐句切分及 segments 数组逻辑，统一为整段音频合成。
CapCut 字幕切分由组装阶段按字符比例分配，简化音频上传、
时间线构建和字幕生成流程，减少冗余处理分支。

.claude/skills/video-from-script/scripts/lib/pipeline-utils.js

@@ -165,8 +165,8 @@ function getManifestDir(manifestPath) {
 // ============================================================================
 
 function splitTextIntoSentences(text) {
-  const sentenceEnders = /[。！？；]/
-  const clauseEnders = /[，：]/
+  // 在句号、感叹号、分号、逗号处断句——它们是口播语音的天然呼吸点。
+  const sentenceEnders = /[。！；，]/
 
   const sentences = []
   let current = ''
@@ -175,16 +175,13 @@ function splitTextIntoSentences(text) {
     current += char
 
     if (sentenceEnders.test(char)) {
-      sentences.push(current.trim().replace(/[。！？；，：、]/g, ''))
-      current = ''
-    } else if (clauseEnders.test(char) && current.length > 8) {
-      sentences.push(current.trim().replace(/[。！？；，：、]/g, ''))
+      sentences.push(current.trim().replace(/[。！；，：？、——…]/g, ''))
       current = ''
     }
   }
 
   if (current.trim()) {
-    sentences.push(current.trim().replace(/[。！？；，：、]/g, ''))
+    sentences.push(current.trim().replace(/[。！；，：？、——…]/g, ''))
   }
 
   return sentences