fix: sanitize message history to prevent API errors

[paulbettner]

Summary

• Add sanitizeMessages() function to strip reasoning parts and empty messages from provider input
• Update abort error filtering to only include messages with contentful parts (text or completed/errored tools)
• Prevents OpenAI API errors when reasoning parts appear in conversation history

Changes

1. transform.ts: Added sanitizeMessages() function that filters out:

   • Messages with only reasoning content type
   • Empty string messages
   • Messages with empty content arrays

2. prompt.ts: Improved abort error handling to use hasContentfulPart logic:

   • Only includes aborted messages that have text parts
   • Or tool parts with completed/error status
   • Excludes messages with only step-start or reasoning parts

Test plan

• Verify that conversations with reasoning parts don't cause API errors
• Confirm that aborted messages with meaningful content are still included in history
• Test that empty or reasoning-only messages are filtered out correctly

🤖 Generated with Claude Code

[paulbettner]

Closing obsolete PR - changes already on dev branch through other merges