…ally and saving openai-compatible messages

Resolved conflicts by accepting main's refactored architecture:
- Removed old flat client files (sync_client.py, async_client.py)
- Accepted new package structure (client/, task/ directories)
- Kept main's BaseClient and BaseTask implementations

Next steps: Port OpenAI v2/message features to the new architecture

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>

Major changes:
- Updated models: Added UploadFileResponse, added raw_output to LLMResponse, removed current_step
- Changed endpoint from /v1/message to /v2/message across all clients
- Added S3 upload functionality to SyncClient and AsyncClient via put_s3_presigned_url()
- Updated BaseClient._build_payload() to use OpenAI-compatible messages_history format
- Updated create_message() to upload screenshots to S3 and build OpenAI-compatible messages
- Updated BaseTask to use client-side UUID generation for task_id
- Updated BaseTask to track message_history instead of last_task_id/history_steps
- Updated Task and AsyncTask to use new v2 API flow (no API call in init_task)
- Updated ShortTask and AsyncShortTask signatures (removed last_task_id/history_steps)

All changes maintain the base class architecture from main branch while
adding OpenAI v2 compatibility features.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>