Fix TF export for BottleneckCSP layer

[nrupatunga]

The current implementation of BottleneckCSP in common.py uses SiLU activation, whereas the tf.py script for exporting uses ReLU, This PR fixes the same.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improved activation function in TensorFlow models within the YOLOv5 repository.

📊 Key Changes

• Changed activation function from ReLU (Rectified Linear Unit) with alpha=0.1 to Swish in TensorFlow models.

🎯 Purpose & Impact

• The purpose of this change is to enhance the performance of the neural network by using the Swish activation function, which has shown better performance in some scenarios compared to ReLU.
• Users may experience improved model accuracy and training stability.
• 🔍 Swish is known to be smoother and can help mitigate vanishing gradient problems, leading to potentially better convergence properties during model training.

[glenn-jocher]

@nrupatunga good catch! PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐