refactor(workflow): 隔离逻辑运行键为 node_group_id

将 checkpoint、异步任务与段结果缓存的唯一键从 execution_id 改为 node_group_id，区分换参重跑与续跑语义；恢复/续跑复用执行记录的组标识，换参重跑则换新组并软删旧组残留，消除软删墓碑导致同键重存失效的问题。

workflow/model/entity/flow_segment_result.go

@@ -6,7 +6,10 @@ import "gitea.redpowerfuture.com/red-future/common/beans"
 type FlowSegmentResult struct {
 	beans.SQLBaseDO `orm:",inherit"` // 嵌入基础字段：Id, TenantId, Creator, CreatedAt, Updater, UpdatedAt, DeletedAt
 	// 业务字段
-	ExecutionId  int64  `orm:"execution_id" json:"executionId" description:"执行ID"`
+	// NodeGroupId 所属逻辑运行(attempt)标识：段结果唯一键 (node_group_id, node_id, segment_index)，
+	// 换参重跑换组即换键，绝不复用被软删墓碑的旧组行（软删即终态，不复活）
+	NodeGroupId  string `orm:"node_group_id" json:"nodeGroupId" description:"所属逻辑运行(node_group_id)标识"`
+	ExecutionId  int64  `orm:"execution_id" json:"executionId" description:"执行ID（溯源，不参与唯一键）"`
 	NodeId       string `orm:"node_id" json:"nodeId" description:"视频生成节点ID"`
 	SegmentIndex int    `orm:"segment_index" json:"segmentIndex" description:"段序号"`
 	VideoKey     string `orm:"video_key" json:"videoKey" description:"视频输出字段key"`
@@ -21,6 +24,7 @@ type SegmentRef struct {
 
 type flowSegmentResultCol struct {
 	beans.SQLBaseCol
+	NodeGroupId  string
 	ExecutionId  string
 	NodeId       string
 	SegmentIndex string
@@ -30,6 +34,7 @@ type flowSegmentResultCol struct {
 
 var FlowSegmentResultCol = flowSegmentResultCol{
 	SQLBaseCol:   beans.DefSQLBaseCol,
+	NodeGroupId:  "node_group_id",
 	ExecutionId:  "execution_id",
 	NodeId:       "node_id",
 	SegmentIndex: "segment_index",