Enable prediction and fine-tuning

[zhiqwang]

🚀 Feature

Use the strategy like lightning-flash to predict and support fine-tuning the models.

Motivation

A way to make the current trainer clean and easy to use.

Pitch

• Predict like lightning-flash #62
• Add mAP Metrics and validation #89
• Add mAP metrics for VOC datasets
• Add validation methods in tasks #93
• Migrate data pipeline to lightning-flash #101
• Cleanup data pipeline #103
• Fixing data augmentation #115
• Add module-frozen mechanism for fine-tuning, refer here and here
• Finetuning: doc link