[BUG] Invalid transcript written if claude sent SIGTERM too soon

Environment

• Platform (select one):
• [x] Anthropic API
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other: <!-- specify -->
• Claude CLI version: 1.0.92 (Claude Code)
• Operating System: 15.5 (24F74)
• Terminal: Spawned as a child process without shell

Bug Description

After sending SIGINT to terminate the process "too early", an invalid transcript can be written out.

This results in the session not being found when we try to resume it again.

The transcript looks as follows:

{"type":"summary","summary":"Megathinking: Collaborative Problem-Solving Strategy Development","leafUuid":"a254249a-5b4f-4fd5-bd10-a846027fc0ff"}

When we try to resume it, CC exits with a non zero exit code and the following message to STDERR: No conversation found with session ID: b008fd78-117e-4c95-bc96-3f281aa84b48

The transcripts firstly should never be invalid (unresumable), and furthermore, they should likely immediately copy over the old history when you resume a session.