Validate task type before model upload

[leeclemnet]

Summary

Uploading a seg .pt with model_type="yolov11" used to succeed silently then crash at inference. Same for rfdetr detect/seg mismatches. Adds two client-side checks:

1. .pt ↔ model_type reconciliation in model_processor.process():
   • YOLO: detects task from Ultralytics class name. Auto-appends the missing suffix when user omitted it; raises on conflict.
   • RF-DETR: detects task from checkpoint["model_name"] (same field rf-detr's loader uses). Raises on conflict.
2. model_type ↔ project type cross-check in Version.deploy().

Also adds canonical TASK_* constants to config.py.

Test plan

• python -m unittest — all pass
• ruff check roboflow — clean
• YOLOv11l-seg .pt + model_type="yolov11" on seg project → deploys as yolov11-seg, inference returns masks
• Same .pt + model_type="yolov11-pose" → ValueError
• model_type="yolov11-seg" on detection project → ValueError
• Regression: detection .pt + model_type="yolov11" on detection project → clean deploy
• rf-detr seg .pt + model_type="rfdetr-medium" → ValueError
• model_type="rfdetr-seg-medium" on detection project → ValueError

🤖 Generated with Claude Code