fix(run-engine): Preserve snapshot checkpoint ID when a PENDING_EXECUTING snapshot stalls (#2493)

Author: ericallam

internal-packages/run-engine/src/engine/index.ts

@@ -1251,6 +1251,7 @@ export class RunEngine {
             },
             orgId: latestSnapshot.organizationId,
             projectId: latestSnapshot.projectId,
+            checkpointId: latestSnapshot.checkpointId ?? undefined,
             error: {
               type: "INTERNAL_ERROR",
               code: "TASK_RUN_DEQUEUED_MAX_RETRIES",

internal-packages/run-engine/src/engine/systems/runAttemptSystem.ts

@@ -1129,6 +1129,7 @@ export class RunAttemptSystem {
     error,
     workerId,
     runnerId,
+    checkpointId,
     tx,
   }: {
     run: TaskRun;
@@ -1142,6 +1143,7 @@ export class RunAttemptSystem {
     error: TaskRunInternalError;
     workerId?: string;
     runnerId?: string;
+    checkpointId?: string;
     tx?: PrismaClientOrTransaction;
   }): Promise<{ wasRequeued: boolean } & ExecutionResult> {
     const prisma = tx ?? this.$.prisma;
@@ -1189,6 +1191,7 @@ export class RunAttemptSystem {
         organizationId: orgId,
         workerId,
         runnerId,
+        checkpointId,
       });
 
       return {