Validate best.pt on train end

[glenn-jocher]

Implements a final validation run (with plots) on best.pt after training completes. Resolves issues raised in #4858.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Simplification of validation logic and cleanup in training end routines.

📊 Key Changes

• Removed verbosity and JSON saving conditions tied to the COCO dataset and final epoch in train.py.
• Validation now always runs without plots unless the best model is being validated.
• Consolidated the optimizer stripping code for the last and best model files.
• In 'val.py', streamlined conditional model precision setting with a single line.

🎯 Purpose & Impact

• The update aims to reduce complexity within the training code by eliminating conditional statements related to verbose flags and JSON outputs.
• This change may lead to a slight performance improvement and better understandability of the code for developers.
• For users, these changes ensure a more consistent behavior at the end of training, regardless of dataset and epoch count.
• Streamlining the precision setting in val.py impacts how the model is run during validation, potentially affecting the speed and resource usage for this step.