Update eval: hybrid sizing gives 0.3% diameter error

Hybrid sizing: detect@0.3 + measure@0.5 core — fixes 10% size bias

Update eval results with CC segmentation @ threshold 0.3

Fix segmentation: CC instead of watershed, threshold 0.3

Add inference.py — bubcount-student CLI for local visualization

Upload bubble_model.onnx with huggingface_hub

Upload README.md with huggingface_hub

Upload eval_results.json with huggingface_hub

Upload history.json with huggingface_hub

Upload config.json with huggingface_hub

Upload final_model.pt with huggingface_hub

Upload best_model.pt with huggingface_hub

initial commit