Add WandB integration and new metrics

[ArgoHA]

Adding Weights and biases integration and metrics: F1-score, Precision, Recall, TPs, FPs, FNs, box IoU. New metrics are calculated with a fixed threshold.

• In runtime.yaml added configs for WandB
• Wandb Logs mAP and new metrics, LR, training loss and system info (GPU usage, vRAM, CPU usege...)
• WandB is not required, if user has it installed - it will be used.
• New class to calculate all new metrics.

I was training D-FINE on 4 datasets, results are great, but I missed comfortable things like WandB and F1-score, hence this PR. People should pay less attention to mAPs when working on a real world tasks, as mAP is calculated with different conf thresholds, but on prod we always fix conf threshold, so it's good to know real world metrics like F1-score.